Add clientSubs, clientSndQ, clientMsgQ, clientThreads counts
and GHC RTS large objects, compact, fragmentation metrics
to identify memory growth not explained by server-level maps.
* smp: deliver service subscription to correct client
* tests: more resilient to concurrency
* optimize PostgreSQL query
* fix service re-association after server "downgrade"
* correctly handle service removed from server (and ID changed)
* remove unused
---------
Co-authored-by: Evgeny @ SimpleX Chat <259188159+evgeny-simplex@users.noreply.github.com>
* xftp-server: embed file download widget in XFTP server web page
When a URL has a hash fragment (>50 chars), the server page shows the
file download UI instead of the server info page. Embeds xftp-web
assets (JS, CSS, crypto worker) and protocol overlay with matching
website content. Overlay renders below the server navbar.
* xftp-server: fix overlay scroll lock, remove extra margin, fix dark SVG
* xftp-server: move file transfer widget to standalone /file page
* web: collapse all repeated Nothing sections in render
section_ only collapsed the first occurrence of a section when content
was Nothing, leaving subsequent sections with the same label intact.
This caused SMP server pages to show raw <x-xftpConfig> tags.
* xftp-server: update bundled css/js
* xftp-server: move file.html to xftp-server, rename xftp bundle dir
* web: remove unused server-info wrapper div
* refactor
* fix
---------
Co-authored-by: Evgeny <evgeny@poberezkin.com>
* web: parameterize generateSite, remove Embedded from library
Move embedFile/embedDir out of the library so it works when
simplexmq is consumed as a dependency. generateSite now accepts
mediaContent, wellKnown, and linkHtml as parameters.
* smp-server, xftp-server: embed static files in executables
Add shared apps/common/Embedded.hs with TH splices, update SMPWeb
and XFTPWeb to pass embedded content to generateSite, move
file-embed dependency from library to executables and test suite.
* refactor
* add export, move common files to Web subfolder
* fix .cabal
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* xftp-web: use XFTP server domain in share link, verify on download
Use a random XFTP server host from the file description as the link
origin instead of the current page domain. On download, verify
that the hosting domain matches one of the servers in the description.
* xftp-web: use first description server as link origin
* xftp-web: show wrong server error in download UI instead of blank page
* xftp: add web page for server information
* web: rename XFTP.Web to XFTPWeb, remove XFTP subdirectory
* refactor(xftp): remove storage quota from web page
* refactor
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* web: extract shared web module from smp-server
Move web serving infrastructure (warp, static files, HTML templating)
from apps/smp-server/web/Static.hs into library modules:
- Simplex.Messaging.Server.Web (generic web infra + templating)
- Simplex.Messaging.Server.Web.Embedded (TH-embedded assets)
Move static assets from apps/smp-server/static/ to
src/Simplex/Messaging/Server/Web/.
Move EmbeddedWebParams/WebHttpsParams from Server.Main to Server.Web.
Keep SMP-specific rendering (serverInformation) in apps/smp-server/SMP/Web.hs.
generateSite is now generic: takes pre-rendered HTML + link page paths,
enabling reuse by XFTP and NTF servers.
* web: add tests for templating engine
Tests for render, section_, item_, and timedTTLText functions
in Simplex.Messaging.Server.Web module.
* web: add serverInfoSubsts, serveStaticPageH2, safe port parsing
* web: rename SMP.Web to SMPWeb, remove SMP subdirectory
* fix(web): section_ collapsing sections with Just "" content
Commit e48bedea ("servers: fix server pages when source code is not
specified") changed section_ to treat Just "" the same as Nothing -
collapsing the section. The intent was to handle the sourceCode case
(empty string when not specified), but the guard
`not (B.null content)` also broke operator, admin, complaints, and
hosting - all of which legitimately use Just "" as a
section-present marker.
Before (correct):
Nothing -> before <> next
Just content -> before <> item_ label content inside <> ...
After (broken):
Just content | not (B.null content) -> ...
_ -> before <> next
Restore the original behavior: only Nothing collapses a section.
* refactor
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* xftp: implementation of XFTP client as web page (rfc, low level functions)
* protocol, file descriptions, more cryptogrpahy, handshake encoding, etc.
* xftp server changes to support web slients: SNI-based certificate choice, CORS headers, OPTIONS request
* web handshake
* test for xftp web handshake
* xftp-web client functions, fix transmission encoding
* support description "redirect" in agent.ts and cross-platform compatibility tests (Haskell <> TypeScript)
* rfc: web transport
* client transport abstraction
* browser environment
* persistent client sessions
* move rfcs
* web page plan
* improve plan
* webpage implementation (not tested)
* fix test
* fix test 2
* fix test 3
* fixes and page test plan
* allow sending xftp client hello after handshake - for web clients that dont know if established connection exists
* page tests pass
* concurrent and padded hellos in the server
* update TS client to pad hellos
* fix tests
* preview:local
* local preview over https
* fixed https in the test page
* web test cert fixtures
* debug logging in web page and server
* remove debug logging in server/browser, run preview xftp server via cabal run to ensure the latest code is used
* debug logging for page sessions
* add plan
* improve error handling, handle browser reconnections/re-handshake
* fix
* debugging
* opfs fallback
* delete test screenshot
* xftp CLI to support link
* fix encoding for XFTPServerHandshake
* support redirect file descriptions in xftp CLI receive
* refactor CLI redirect
* xftp-web: fixes and multi-server upload (#1714)
* fix: await sodium.ready in crypto/keys.ts (+ digest.ts StateAddress cast)
* multi-server parallel upload, remove pickRandomServer
* fix worker message race: wait for ready signal before posting messages
* suppress vite build warnings: emptyOutDir, externals, chunkSizeWarningLimit
* fix Haskell web tests: use agent+server API, wrap server in array, suppress debug logs
* remove dead APIs: un-export connectXFTP, delete closeXFTP
* fix TypeScript errors in check:web (#1716)
- client.ts: cast globalThis.process to any for browser tsconfig,
suppress node:http2 import, use any for Buffer/chunks, cast fetch body
- crypto.worker.ts: cast sha512_init() return to StateAddress
* fix: serialize worker message processing to prevent OPFS handle race
async onmessage allows interleaved execution at await points.
When downloadFileRaw fetches chunks from multiple servers in parallel,
concurrent handleDecryptAndStore calls both see downloadWriteHandle
as null and race on createSyncAccessHandle for the same file,
causing intermittent NoModificationAllowedError.
Chain message handlers on a promise queue so each runs to completion
before the next starts.
* xftp-web: prepare for npm publishing (#1715)
* prepare package.json for npm publishing
Remove private flag, add description/license/repository/publishConfig,
rename postinstall to pretest, add prepublishOnly, set files and main.
* stable output filenames in production build
* fix repository url format, expand files array
* embeddable component: scoped CSS, dark mode, i18n, events, share
- worker output to assets/ for single-directory deployment
- scoped all CSS under #app, removed global resets
- dark mode via .dark ancestor class
- progress ring reads colors from CSS custom properties
- i18n via window.__XFTP_I18N__ with t() helper
- configurable mount element via data-xftp-app attribute
- optional hashchange listener (data-no-hashchange)
- completion events: xftp:upload-complete, xftp:download-complete
- enhanced file-too-large error mentioning SimpleX app
- native share button via navigator.share
* deferred init and runtime server configuration
- data-defer-init attribute skips auto-initialization
- window.__XFTP_SERVERS__ overrides baked-in server list
* use relative base path for relocatable build output
* xftp-web: retry resets to default state, use innerHTML for errors
* xftp-web: only enter download mode for valid XFTP URIs in hash
* xftp-web: render UI before WASM is ready
Move sodium.ready await after UI initialization so the upload/download
interface appears instantly. WASM is only needed when user triggers
an actual upload or download. Dispatch xftp:ready event once WASM loads.
* xftp-web: CLS placeholder HTML and embedder CSS selectors
Add placeholder HTML to index.html so the page renders a styled card
before JS executes, preventing layout shift. Use a <template> element
with an inline script to swap to the download placeholder when the URL
hash indicates a file download. Auto-compute CSP SHA-256 hashes for
inline scripts in the vite build plugin.
Change all CSS selectors from #app to :is(#app, [data-xftp-app]) so
styles apply when the widget is embedded with data-xftp-app attribute.
* xftp-web: progress ring overhaul
Rewrite progress ring with smooth lerp animation, green checkmark on
completion, theme reactivity via MutationObserver, and per-phase color
variables (encrypt/upload/download/decrypt).
Show honest per-phase progress: each phase animates 0-100% independently
with a ring color change between phases. Add decrypt progress callback
from the web worker so the decryption phase tracks real chunk processing
instead of showing an indeterminate spinner.
Snap immediately on phase reset (0) and completion (1) to avoid
lingering partial progress. Clean up animation and observers via
destroy() in finally blocks.
* xftp-web: single progress ring for upload, simplify ring color
* xftp-web: single progress ring for download
* feat(xftp-web): granular progress for encrypt/decrypt phases
Add byte-level progress callbacks to encryptFile, decryptChunks,
and sha512Streaming by processing data in 256KB segments. Worker
reports fine-grained progress across all phases (encrypt+hash+write
for upload, read+hash+decrypt for download). Progress ring gains
fillTo method for smooth ease-out animation during minimum display
delays. Encrypt/decrypt phases fill their weighted regions (0-15%
and 85-99%) with real callbacks, with fillTo covering remaining
time when work finishes under the 1s minimum for files >= 100KB.
* rename package
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
---------
Co-authored-by: Evgeny @ SimpleX Chat <259188159+evgeny-simplex@users.noreply.github.com>
Co-authored-by: shum <github.shum@liber.li>
Co-authored-by: sh <37271604+shumvgolove@users.noreply.github.com>
* agent: use strict tables
* migrate existing tables to strict
* test: verify that all tables are strict
* fix column types for device_token and ntf_mode
* fix encodings and column types for ntf_sub_action and ntf_sub_smp_action
* update schema
* remove debug.trace
* log
* agent: remove service queue association when service ID changed
* agent: process ENDS event
* agent: send service subscription error event
* agent: test migrating to/from service subscriptions, fixes
* agent: always remove service when disabled, fix service subscriptions
* agent: remove service/queue associations when service ID changes
* agent: check that service ID in NEW response matches session ID in transport session
* agent subscription WIP
* test
* comment
* enable tests
* update queries
* agent: option to add SQLite aggregates to DB connection (#1673)
* agent: add build_relations_vector function to sqlite
* update aggregate
* use static aggregate
* remove relations
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* add test, treat BAD_SERVICE as temp error, only remove queue associations on service errors
* add packZipWith for backward compatibility with GHC 8.10.7
---------
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* agent: use server keyhash when loading service record
* agent: process queue/service associations with delayed subscription results
* agent: service subscription events
* servers: maintain xor-hash of all associated queue IDs in PostgreSQL (#1615)
* ntf server: maintain xor-hash of all associated queue IDs via PostgreSQL triggers
* smp server: xor hash with triggers
* fix sql and using pgcrypto extension in tests
* track counts and hashes in smp/ntf servers via triggers, smp server stats for service subscription, update SMP protocol to pass expected count and hash in SSUB/NSSUB commands
* agent migrations with functions/triggers
* remove agent triggers
* try tracking service subs in the agent (WIP, does not compile)
* Revert "try tracking service subs in the agent (WIP, does not compile)"
This reverts commit 59e908100d.
* comment
* agent database triggers
* service subscriptions in the client
* test / fix client services
* update schema
* fix postgres migration
* update schema
* move schema test to the end
* use static function with SQLite to avoid dynamic wrapper
* smp server: refactor message delivery to always respond SOK to subscriptions
* refactor ntf subscribe
* cancel subscription thread and reduce service subscription count when queue is deleted
* subscribe rcv service, deliver sent messages to subscribed service
* subscribe rcv service to messages (TODO delivery on subscription)
* WIP
* efficient initial delivery of messages to subscribed service
* test: delivery to client with service certificate
* test: upgrade/downgrade to/from service subscriptions
* remove service association from agent API, add per-user flag to use the service
* agent client (WIP)
* service certificates in the client
* rfc about drift detection, and SALL to mark end of message delivery
* fix test
* fix test
* add function for postgresql message storage
* update migration
* agent: make createConnection and setConnShortLink apis support setting all link data fields
* add functions
* refactor
* refactor
* fix tests
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* agent: support client notices
* improve
* fix, test
* rename
* cleanup
* send and process notices in more cases
* dont delete
* dont remove notice on other permanent errors
* dont remove notice if there is no notice ID in queue
* add server to error
* allow deleting
* only use notice if key hash matches
* agent: subscribe all connections
* query, version
* BoolInt
* add query to errors
* Revert "add query to errors"
This reverts commit 32a1f7fe11.
* fix optional field
* version
* limit number of in-flight subscriptions to 35000
* agent: optimize subscriptions memory usage more (do not store subscribed queues in memory) WIP
* use new session subscriptions data
* version
* remove old data structure
* remove version
* batch deletions
* test TSessionSubs
* comment
* smp server: limit by time the queues to export journal messages for
* pass queue/msg thresholds separately
* reset db connection on errors
* Revert "smp server: limit by time the queues to export journal messages for"
This reverts commit d3bc0cba4b.
* fix test compilation
* flag to expire messages
* improve test
* expire messages newer than quota
* smp server: store messages in PostgreSQL
* stored procedures to write and to expire messages
* function to export messages
* move all message functions to PostgreSQL, remove delete trigger
* comments
* import messages to db
* fix message import, add export
* fix export
* fix export
* fix compilation flags
* import messages line by line
* fix server start with database storage
* fix compilation
* comments
* smp server: fix server pages when source code is not specified
* servers: include git commit in version
* flexible alpn
* fix test
* fix ghc 8.10.7 build
* servers: prohibit changing role during control port session
* quota for blocked queues
* allow disabling blocking and quota
* fix test
* fix INI file
* smp protocol: create notification credentials via NEW command that creates the queue
* create ntf subscription for queues created with ntf credetials
* do not create ntf credentials when switching connection to another queue
* smp server: refactor subscriptions and delivery
* metric for time between MSG and ACK
* cleanup
* refactor pattern match for ghc 8.10.7
* time buckets
* split max time metric
* histogram
* fix
* agent: fix message delivery in case one of the connections has no snd queue for any reason - it could break delivery to all connections
* simplify
* comment
* refactor, also postpone failing on ratchet sync send prohibited errors
* postpone failing on connection errors to allow subsequent connections succeed
* don't exclude postgres modules for client_library flag
* fix build with client and server postgres flag
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* time buckets
* split max time metric
* histogram
* histogram for confirmed delivery times
* gaugehistogram
* fix created, _ in gauge_histogram
* remove comments
* fix metrics
* protocol: refactor types and encoding
* clean
* smp server: batch commands (#1560)
* smp server: batch commands verification into one DB transaction
* ghc 8.10.7
* flatten transmission tuples
* diff
* only use batch logic if there is more than one transmission
* func
* reset NTF service when adding notifier
* version
* Revert "smp server: use separate database pool for reading queues and creating service records (#1561)"
This reverts commit 3df2425162.
* version
* Revert "version"
This reverts commit d80a6b74c5.
* agent: use PQ keys in contact request data inside link container (but not in contact request link); use PQ keys in invitations sent to contact addresses
* do not use PQ keys in the link with old address versions
* log alpn
* always use HTTP when SNI is sent, regardless of ALPN
* decide credential based on SNI
* do not use web port in SMP/Ntf servers connecting to SMP servers
* simpler
* refactor
* fix
* rfc: client certificates for high volume clients (opertors' chat relays, notification servers, service bots)
* client certificates types (WIP)
* parameterize Transport
* protocol/schema/api changes
* agent API
* rename command
* agent subscriptions return local ClientServiceId to chat
* verify transmissions
* fix receiving client certificates, refactor
* ntf server: remove shared queue for all notification subscriptions (#1543)
* ntf server: remove shared queue for all notification subscriptions
* wait for subscriber with timeout
* safer
* refactor
* log
* remove unused
* WIP service subscriptions and associations, refactor
* process service subscriptions
* rename
* simplify switching subscriptions
* SMP service handshake with additional server handshake response
* notification delivery and STM persistence for services
* smp server: database storage, store log, fix encoding for STORE error, replace String with Text in locks and error
* stats
* more stats
* rename SMP commands
* service subscriptions in ntf server agent (tests fail)
* fix
* refactor
* exports
* subscribe ntf server as service for associated queues
* test ntf service connection, fix SOKS response, fix service associations not removed in STM storage
* INI option to support services
* ntf server: downgrade subscriptions when service is no longer supported, track counts of subscribed queues
* smp protocol: include service certificate fingerprint in the string signed over with entity key (TODO two tests fail)
* fix test
* ntf server prometheus stats, use Int64 in SOKS/ENDS responses (to avoid conversions), additional error status for ntf subscription
* update RFC
* refactor useServiceAuth to avoid ad hoc decisions about which commands use service signatures, and to prohibit service signatures on other commands
* remove duplicate service signature syntax check from checkCredentials, it is checked in verifyTransmission
* service errors, todos
* fix checkCredentials in ntf server, service errors
* refactor service auth
* refactor
* service agent: store returned queue count instead of expected
* refactor serverThread
* refactor serviceSig
* rename
* refactor, rename, test repeat NSUB service association
* respond with error to SUBS
* smp server: export/import service records between database and store log
* comment
* comments
* ghc 8.10.7
* agent: setInvitationShortLink api
* Eq instance
* allow changing link data on server, refactor
* fix
* encodings
* remove link data after connection
* Revert "encodings"
This reverts commit f8e254cca9.
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* parameterize transport by peer type (client/server)
* LogDebug level when test is retried
* support "flipped" HTTP2, fix test retry to avoid retrying pending tests
* move sync to the end of the tests
* agent: return error and message absence differently when getting notification messages
* fix test
* mapM
* inline nse functions, release lock on error or no message
* agent: handle cases when last message ts is not set for notifications; set last ts for "stale" notifications when messages expired and queue is empty, to prevent repeated processing
* only log errors if they exist
* only set last ts for queue that delivered notification
* ntf server, agent: send all periodic notifications from one thread, only to old active clients or new clients with periodic notification mode
* send different type via subscription queues
* option to compact store log on start
* ntf server: skip duplicates when importing tokens and subscriptions
* skip imported last notifications when no token or subscription present
* fix skipped imported notifications count
* all tests
* fix test
* ntf server: allow retries when creating subscriptions, prohibit subscriptions with the same queue but another notifier key or token
* sync files in the test
* refactor
* agent: option to use web port by default for preset servers only
* shorten/restore short links in agent, add encodings for SMP web port setting
* decouple preset domains from preset servers for short links
* refactor, rename
* smp server: short links and owners for channels
* types
* support mutliple rcv keys
* fix down migration, test/create server schema dump
* reduce schema dump
* parameterize type for link data by connection type
* return full connection link data
* test version
* change short link encoding
* test: print pg_dump output
* server pages, link encoding
* fix connection request when queue data and sender ID are created for old servers
* test, change pattern
* ci: install postgresql tools in runner (#1507)
* ci: install postgresql tools in runner
* ci: docker shell abort on error
* fix pattern for ghc 8.10.7
* patch ConnReqUriData SMP encoding to preserve queue mode after decoding
* test for RKEY
* fix/test store log with RKEY
---------
Co-authored-by: sh <37271604+shumvgolove@users.noreply.github.com>
* agent: padded encryption for link data, tests
* lambda
* test short links via proxy
* tests: server persistence with short links
* rfc: group links
* shorten, restore, test short links encoding
* rfc
* agent: join connection when 1-time invitation short link is already secured
* do not pass short link to join
* delete short link record after connection
* smp server: remove locks for deleted queues, additional statistics for objects in memory
* version
* reduce queue cache usage
* less caching, refactor
* comments
* revert version
* smp protocol: short links types and other changes from RFC
* add fields for queue link ID and data
* create queue and ntf credentials with NEW command
* all tests
* simplfiy types, update rfc
* update rfc
* include SenderId in NEW request in case queue data is sent
* store queue data and generate link ID if needed
* update rfc
* agent API and types
* SMP commands and persistence for short links
* SMP client functions for short links
* agent client functions for short links
* create rcv queue with short link (TODO secret_box)
* encryption and encoding for link data, postgres client migration
* test creating short link
* get link and data, tests
* comments
* type signature
* smp server: use COPY to import store log to postgres db
* compact queues when importing to postgres
* mempty
* version
* handle errors while expiring, mask async exceptions while getting queue
* whitespace
* version
* smp server: split postgres support to a separate executable, to not require postgres library in the main binary
* comments
* enable server_postgres flag by default, add CPP option to test
* refactor
* change default for server_postgres to False
* diff
* smp server: expire only active queues
* version
* do not cache all queues while processing expirations
* refactor
* foldWithOptions_
* version
* use shared lock when expiring all queues
* use TMVar
* comment
* rename
* remove fold options
* do not create locks in the Map for temporarily loaded queues
* fix
* revert version
* smp server: optionally maintain store log with postgres storage (without loading and compacting, for debugging during migration)
* refactor
* remove comment
* smp server: queue store typeclass
* parameterize JournalMsgStore
* typeclass for queue store
* postgres WIP
* compiles, passes tests
* remove StoreType
* split migrations
* progress
* addQueueRec
* reduce type spaghetti
* remove addQueue from typeclass definition
* getQueue
* test postgres storage in SMP server
* fix schema
* comment
* import queues to postgresql
* import queues to postgresql
* log
* fix test
* counts
* ci: test smp server with postgres backend (#1463)
* ci: test smp server with postgres backend
* postgres service
* attempt
* attempt
* empty
* empty
* PGHOST attempt
* PGHOST + softlink attempt
* only softlink attempt
* working attempt (PGHOST)
* remove env var
* empty
* do not start server without DB schema, do not import when schema exists
* export database
* enable all tests, disable two tests
* option for migration confirmation
* comments
---------
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* smp server: remove empty journals when opening message queue
* update, do not backup state
* test
* version
* do not close queue state when queue is opened for writing
* comment
* quota = 4
* refactor openMsgQueue to prevent extra state backups
* use interval in config
* version, expire backups after 5 min
* refactor
* test
* agent: store shared message body only once (when it is the same across messages when batching)
* rename
* refactor
* refactor
* save bodies and messages in single transaction
* comment
* comment
* comment
* box
* mapME
* box
* ValueOrRef
* remove instances
* refactor
* comments
* test
* refactor
* mapAccumLM compatibility with ghc 8.10.7
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* agent: save message body once (plan, schema)
* split
* new type
* bs
* encrypt on delivery
* schema
* fix test
* check pad size
* rename
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* ntf server: record token invalidation reason, add date of the last token activity
* update time
* rename
* optional
* include token ID in delivery error
* version
* protocol version
* fix, log error
* xftp server: use recipient ID in control port to delete and block files
* cap smp proxy agent version at 10
* version
* fix prometheus
* fix
* remove old version support
* log connection parameter on error
* tests
* log sent command tag
* log error and client version
* cap proxy version for previous destination server
* comment, test
* remove logging tag
* remove logs
* version
* SMP version 14
* version
* remove comments
* version
* servers: blocking records for content moderation
* update
* encode BLOCKED as AUTH in old versions
* update
* unblock queue command
* test, status command
* server: support server roles and operators
* make server operator optional
* allRoles
* fix test
* different server host in tests
* remove ServerCfg fields used only in UI
* comments
* choose different server for invitation when connecting via address
* fix test in ghc8107
* simplify
* smp server: do not open/read journal message queues that are known to be empty
* cleanup
* version
* close empty queues on first subscription
* revert version
* smp server: combine queue and message store into one class (WIP)
* keep deleted queue tombstones to prevent race conditions and errors when restoring
* move store log from server to store implementations
* STMQueueStore type class
* fix store closed when messages expired, handle store writing errors
* types
* version
* fix recovery from missing write journal, tests
* version
* ntf: get messages for multiple last notifications (#1352)
* ntf: separate get ntf conns api (#1379)
* ntf: separate get ntf conns api
* nonempty
* update
* update
* remove single get api
* fix test
* refactor
* refactor
* ntf: batch get connections (#1387)
* ntf: batch get apis
* works
* fix
* fix
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* smp server: fixed logging format for journal store errors
* version
* colon
* logs
* refactor
* space
* remove comment
* log file name in fixFileSize
* logError
* stricter
* process all queues more efficiently
* use monoid for queue processing
* expire messages concurrently
* concurrently 2
* Revert "concurrently 2"
This reverts commit c1aee1f22c.
* Revert "expire messages concurrently"
This reverts commit fc53137cdb.
* show queue directory or ID in errors
* foldM
* mask_
* try
* mask more
* refactor
* command to delete journal
* uninterruptibleMask_ when writing to state file
* fix ghc8.10.7
* version
* revert version
* smp server: journal store queue state validation
* validate state on update, refactor
* handle message parsing errors
* typo
* fix test, throw exception when file is smaller than expected
* fix
* fix test in ghc 8.10.7
* core: updated journal store API
* parameterize JournalState by read/write type
* collect stats when importing/expiring
* compare stored stats and store for messages and notifications
* simplify
* smp server: remove STM function from MsgStore
* polymorphic MsgStore
* jourmal storage for messages (WIP)
* more journal, test setup
* writeMsg
* test
* tryDelMsg
* delMsgQueue
* remove MsgStoreClass instance of existential wrapper for Msg stores
* store config
* extract common logic out of store instances
* add store type to config
* open journals, cache last message, tests pass
* CLI commands
* refactor import/export messages
* cli commands to import/export journal message store
* export journal without draining, import/export tests
* journal command
* import/export progress
* better progress info
* only log queue state once when importing
* logs
* handle IO errors in journal store, return as STORE error
* recover from state file errors
* fix message files after crash
* fix messages folder
* .401
* stats for undelivered notifications
* logs, stats
* control port show ntf client IDs
* check that Ntf client is still current and that queue is not full, drop notifications otherwise
* prevent losing notifications when client is not current or queue full
* add log when no notifications, remove some logs
* reduce STM transaction
* revert version change
* smp server: pass server information via CLI during server initialization
* more info
* enable client expiration by default, disable port 8000
* update
* ntf server: control port
* version .405
* control
* use own_server_domains from INI file
* fix subs by server in control port
* bigger queues
* ntf server: only print subscriptions per own server when they are > 0
* fix tests
* revert version change
* dont import listThreads in ghc 8.10.7
* smp-server: Allow serving HTTPS and transport on the same port
* update rfc
* servers: refactor TLS credentials
* provide server credentials in SNI hook
* determine TLS server params dynamically, when starting the server
* remove alpn from TransportServerConfig to decide it dynamically where server is started
* diff
* combine HTTP and SMP on the shared port
* Update to SockAddr
* Fix params and web.https parser
* Switch fork urls
* WIP: add smpServerTestStatic test
* Update warp-tls repo
* shared connection tests
* cleanup
* Add protocol tests
* rename cert file, enable both ports and web by default
* terminate with message on missing credentials
* test cert file
* client option to use port 443 as default SMP port
* use SNI in non-SMP clients
* supported
* remove TODO
* advice
* fix test build
* Add RSA-4096 check for web creds, fix test
* Remove directory listing from static app
* message
* messages
* update log tests
---------
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* servers: refactor TLS credentials
* provide server credentials in SNI hook
* determine TLS server params dynamically, when starting the server
* remove alpn from TransportServerConfig to decide it dynamically where server is started
* agent: transport isolation mode "Session" (default) to use new SOCKS credentials when client restarts or SOCKS proxy configuration changes
* fix test
* agent: support socks proxy without isolate-by-auth, with and without credentials
* add unit tests
* make xftp use correct SOCKS credentials
* rename
* support ipv6 in brackets, test parsing
* constant
* textToHostMode
* space
* agent: don't prohibit deletion of notifications token if different is passed (fixes turning notifications off after importing database)
* ignore error
* notify
* smp protocol: send DELD when subscribed queue is deleted
* fix, test
* refactor
* send DELD event only if the client supports it (version 10); send END otherwise
* fix test
* notify on notifier rotation
* increase test delays
* smp server: add created/updated/used date to queues to manage expiration, all: make Map updates strict in value
* remove strict
* remove time precision
* diff
* style
* only update when time changed
* smp server: fewer map updates on re-subscriptions
* temp version
* replace Client with ClientId in queues
* version
* version
* comments
* reduce threads when sending ENDs
* revert version
* put DRG state to IORef, split STM transaction of sending notification (#1288)
* put DRG state to IORef, split STM transaction of sending notification
* remove comment
* remove comment
* add comment
* revert version
* newtype for server entity IDs, fix TRcvQueues
* Revert "put DRG state to IORef, split STM transaction of sending notification (#1288)"
This reverts commit 517933d189.
* logServer
* put DRG state to IORef, split STM transaction of sending notification (#1288)
* put DRG state to IORef, split STM transaction of sending notification
* remove comment
* remove comment
* add comment
* revert version
* smp server: get message queue faster, avoiding STM contention if queue exists
* IORef for counter
* Revert "put DRG state to IORef, split STM transaction of sending notification (#1288)"
This reverts commit 517933d189.
* version
* remove IORef
* split notification delivery transations
* revert version
* agent: support new connection user id update
* another way for assertion
* add more tests to setConnUserId
* remove fdescribes
* allow rcv connection to change user id
* add functional test to api
* remove fdescribe
* refactor
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* agent: use weak ThreadId and forkIO in workers instead of async (reduce memory)
* agent: do not start and exit delivery workers when there are no messages to deliver (#1264)
* agent: exit delivery workers when no messages to deliver
* only start delivery workers when there are pending messages
* fix
* focus test
* enable all tests
* lift
* do not exit workers when there is no work
* agent: retry loop that resumes subscriptions as soon as agent is moved to foreground, suspend retry loops while agent is suspended
* reset retry enterval when moving to foreground
* account for network state too
* simplify
* typo
* simplify
* ntf server: only mark subscriptions as pending if the disconnected client is current
* add sessionId to subscribed queue
* add sessionId to subscriptions in ntf server agent
* fix
* agent: fix possible dead lock between sending and receiving messages, stress test for message delivery
* deliver events after the lock is released
* delayed delivery in command processing too
* tests: increase message expiration time
* rfc: faster handshake protocol
* update
* 1 message
* SKEY
* use SKEY for both parties
* test
* update doc
* NEW command parameter
* add k=s param to queue URI
* fix
* add sndSecure field to queues
* make sender key non-optional in SndQueue (WIP, tests fail)
* fast handshake sometimes works (many tests fail)
* correctly handle SKEY retries, avoiding to re-generate the keys
* handle SKEY retries during async connection
* fix most tests (1 test fails)
* remove do
* fix contact requests encoding/tests
* export
* fix: ignore duplicate confirmations, fixes testBatchedPendingMessages
* do not store sndSecure in store log if it is false to allow server downgrade
* add connection invitation encoding tests
* agent: report correct errors from xftp handshake so they are treated as temporary
* disable slow servers test
* remove comments
* all tests
* remove duplicate functions
* agent: report correct errors from xftp handshake so they are treated as temporary
* disable slow servers test
* remove comments
* all tests
* remove duplicate functions
* SMP server information
* fix tests
* country codes
* smp-server: serve contact and link pages from static files (#1084)
* smp-server: serve contact and link pages from static files
* generate index
* use params from ini
* render using ServerInformation
* tweak templates
* update
* fix some html
* smp-server: fix layout (#1097)
* smp-server: fix layout
* port fixes to link page
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* update server information page
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
Co-authored-by: M. Sarmad Qadeer <MSarmadQadeer@gmail.com>
* update server info
* web: improve server info page design (#1166)
* web: improve server info page design
* web: fix font errors & some tags
* web: improve contact & invitation page layout and header
* update
* remove unused files/css
* cleanup
* fix link page
* remove unused font links
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* show contact address as is
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: M. Sarmad Qadeer <MSarmadQadeer@gmail.com>
* smp: put client proxy command processing threads under a shared semaphore
* add LIMITS.max_proc_threads to server config
* rename to PROXY.client_concurrency
* retry on strictly greater than max concurrency
* set default to 16
* rename
* fix limit
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* agent: aggregate multiple expired subscription responses into a single UP event
* clean up
* refactor processing of expired responses
* refactor
* refactor 2
* refactor unexpectedResponse
* smp: use session vars for reconnecting small agent
* process errors
* split session and protocol functions
* add active flag to agent
* actually invoke agent shutdown
* close proxy agent too
* restore stopping ntf subscribers
* agent: do not mark subscriptions on expired sessions as active, do mark delayed subscriptions as active on the same session, SUBOK response in the next SMP protocol version
* client: prevent sub actions from zombie sessions (#1122)
* client: prevent sub actions from zombie sessions
* error handling
* add AERR to pass background errors to client
* switch to activeClientSession
* put closeClient under activeClientSession
* rename
* remove AERR, do not skip processing
* move check and state update to one transaction
* catch extra UPs
* fix
* check queue is still pending before making it active
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* do not forward agent error
* revert not expiring sending subs
* fixes
* track subscription responses better
* add pending connection
* Revert "revert not expiring sending subs"
This reverts commit 4310a69391.
* do not expire sending commands
* rename
* fix race
* function
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* proxy: negotiate client-relay version, include it in PFWD commands and in encrypted forwarded transmissions
* rename
* inline
* comment
* use correct server version when encoding forwarded commands
* proxy: include delivery path in SENT event
* send MWARN event to user on server version or host more errors
* Revert "proxy: include delivery path in SENT event"
This reverts commit 5c476718ec.
* transport: reduce ping traffic
* make pings opt-in, enable automatically with SUB commands
* fix reduced delays
* enable pings on MSG too
* rename pingErrorCount
* check timeout counter even when not sending pings
* clean up
* reset timeout error count on any event
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* SMP proxy: low level client and server implementation
* SMP proxy: server implementation (#1098)
* wip
* PRXY command
* progress
* SMP Proxy: client-level implementation (#1101)
* buildable
* encode messages
* update pkey
* fix queue types
* wrap SEND in proxy lookup
* WIP proxy client
* WIP
* post-rebase fixes
* encode something with something
* cleanup
* update
* fix nonce/corrId in batchingTests
* WIP: dig into createSMPProxySession
* agent
* test progress
* pass the test
* parameterize transport handle with transport peer to include server certificate (#1100)
* parameterize transport handle with transport peer to include server certificate
* include server certificate into THandle
* load server chain and sign key
* fix key type
* fix for 8.10
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* cleanup
* add 2-server test
* remove subsumed test
* checkCredentials for BrokerMsg
* skip batching tests
* remove userId param
* remove agent changes
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* remove unused type
* icrease test timeout
* reduce transport block
* envelope sizes
* don't fork unless have proxied commands to process
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* parameterize transport handle with transport peer to include server certificate
* include server certificate into THandle
* load server chain and sign key
* fix key type
* fix for 8.10
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* parameterize transport handle with transport peer to include server certificate
* include server certificate into THandle
* load server chain and sign key
* fix key type
* fix for 8.10
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* test with failing files (in progress)
* print
* add replica uploading state
* Revert "add replica uploading state"
This reverts commit 7068213aa6.
* <=
* fix
* prints
* test no redundancy
* all tests no redundancy
* revert delay
* refactor
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* log control port commands
* auth
* add auth to xftp, config and commands
* log missing auth
* put smp save under auth
* corrections
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* convert compress to a wrapper with passthrough fallback
* add compress1 for non-batched compression
* use original size as upper bound for scratch
* refactor
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* envelope sizes dependent on PQ encryption (WIP)
* add "supported" flag to ratchets, update this flag on ratchet resync
* change connection PQ status on sendMessage
* comment, fix
* refactor
* fix agent performance leak when re-connecting clients, optimize SMP server
* let the clients remove themselves from their clientvars
* fix another test
* do not call removeSubs when AgentClient is not active
* revert some changes
* revert more, refactor
* comment
* rename
* refactor
* refactor
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* smp: command authorization
* fix encoding, most tests
* remove old tests
* authorize via crypto_box
* extract authenticator to Crypto module
* make TransmissionAuth Maybe
* rfc
* support authenticators in NTF protocol, test matrix (no backwards compatibility yet from new clients to old servers)
* fix/add tests, add version config to "small" agent
* separate client and server versions for SMP protocol
* test batching SMP v7
* do not send session ID in each transmission
* refactor auth verification in the server, split tests
* server "warm up" fixes timing test
* uncomment SUB timing test
* comments, disable two timing tests
* rename version
* increase auth timing test failure threshold
* use different algorithms to authorize snd/rcv commands, use random correlation ID
* transport: fetch and store server certificate (#985)
* THandleParams (WIP, does not compile)
* transport: fetch and store server certificate
* smp: add getOnlinePubKey example to smpClientHandshake
* add server certs and sign authPub
* cleanup
* update
* style
* load server certs from test fixtures
* sign ntf authPubKey
* fix onServerCertificate
* increase delay before sending messages
* require certificate with key in SMP server handshake
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* remove dhSecret from THandle
* remove v8, merge all changes to one version
* parameterize THandle
* rfc: transmission ecnryption
* Revert "parameterize THandle"
This reverts commit 75adfc94fb.
* use batch syntax for ntf server commands
* separate encodeTransmission when there is no key
* typo
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* rename
* diff
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* xftp: add URI encoding for FileDescription
* tweak URI
* allow smaller blocks
* draft xftpReceiveFileFollow' and xftpSendFilePublic'
* add sending with redirect
* allow 64k chunks
* add migrations with redirect fields
* add test case
* fix deadlock
* revert CLI code
* WIP: working send/receive via URI
* fix field ambiguity
* cleanup
* update agent db schema
* update minimal chunk size
* add rfc
* apply suggestions from code review
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* add createRcvFileRedirect
* extract Simplex.Messaging.ServiceScheme and reuse for files
* update db schema
* check size/digest on receive complete
* cleanup
* use SIZE/DIGEST errors for redirects too
* split digest/size errors from redirect checks
* fix redirect error encoding
* rename RedirectMeta to RedirectFileInfo
* use query encoding for file URI
* group maybe fields under RcvFileRedirect
* add extras field
* update rfc
* add extras encoding and no-redirect tests
* fix toStrict for old ghc
* extra client data in file descr URI
* remove decoded yaml file
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* smp-server: add gen-online CLI command
* use CN and algo from old certificate
* add cert checks to test
* rename command
* fix test
* cert
---------
Co-authored-by: Evgeny Poberezkin <evgeny@poberezkin.com>
* tests: add ntf case with multiple Ntf servers
* simplify test
* fix for master
* add server switch test
* add server switch test
* add message test for ntf server switch
* smp-server: check queue balance in stats vs store
* smp-server: add msgExpired stats
* add msgExpired stats
* split expire/stats transactions
* count and pass msgExpired explicitly
* save/load qCount and use it for checking store
* xftp: test file reception - shouldn't get stuck if file is deleted on server
* comment
* expiration test
* approach
* wip
* sort by retries in other works, revert some diff
* revert diff
* modify tests
* refactor
* refactor
* remove prints
* apply to other workers
* remove import
* comment
* refactor
* revert queue size
* fix test
* rename
* comment, correct number of retries
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* control: add delete command
* logDeleteQueue only when found
* use default StrEncoding for CPDelete arg
* move stats update from main transaction
* use size
* stabilize AUTH timing tests
* more iterations
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* agent: mark work items failed (WIP)
* add tests, created_at
* getWorkItem for snd and rcv files
* fix
* tests
* fix
* tests
* test
* tests
* rename
* fix,refactor
* add indexes
* update schema
* do not try to get more work when resuming an existing worker
---------
Co-authored-by: spaced4ndy <8711996+spaced4ndy@users.noreply.github.com>
* agent: notify about polled message processing (for iOS notifications)
* optionally keep key and support re-opening database
* exports
* test that cannot reopen when created with keepKey: false
* set max number of messages to receive for a notification to 3
* Revert "raise mtl version to 2.3.1 (#912)"
This reverts commit 117168ccce.
* Revert "expand dependency ranges (#911)"
This reverts commit 90a8fc91d3.
* fix test
* remove unnecessary forks
* make text version dependent on GHC version
* text
* remove stack.yaml
* bytestring
* bytestring
* hpack 0.35
* mtl
* add sntrup761 source
* it compiles
* Wrap bindings in non-FFI types
Test passes with a dummy RNG.
* pass ChaChaDRG via FunPtr
* Add iOS smoke test at createAgentStore
* style
* add "ssl" library dep
Attempt to fix missing _SHA512 symbol on macos.
* remove sha512 wrapper and use openssl directly
* restore names, remove dummy RNG
* Revert "remove sha512 wrapper and use openssl directly"
This reverts commit f9f7781f09.
* restore code from RFC
* shorter names
* enable all tests
* remove run test
---------
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
* Cut transport server to allow custom tcp servers
Allows socket inspection before wrapping up in a transport/prototocol.
* rename
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* Add X509 cert and TLS credentials generator
* Expand Crypto toolkit and rewrite tls credentials with it
* Exclude X keys from SignatureAlgorithmX509 and TLS.PrivKey
* Add helpers for DB marshalling and fingerprints
* Derive public key from private
* remove module name from selectors
* Remove StrEncoding (PrivateKey Ed25519)
* remove comment
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* Split HTTP2 server and client setup
For attaching to an existing TLS session.
* Add genTlsCredentials
* Allow chain construction from separate credentials
So the CA may be stored and leaf ephemeral.
* Rewrap X509 fingerprint into simplex KeyHash used in transport
* Fix docstring
* Remove TLS.Credentials generator
* Trace auto-subs flag
* Replace Bools with SubscriptionMode
* Handle SMOnlyCreate
* Wire remaining todos
* Update tests and fix
* Bump protocol level
* Apply suggestions from code review
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* Scrub needs_sub from agent DB
* Scrub a few more needSubs from the agent api
* change API, fix test
* agent: do not subscribe to queue when creating reply queue
* fix encoding
* WIP: SMOnlyCreate test
* Add SM guard for confirmQueue
Allows the test case to pump the allowConnection
reply without getting PROHIBITED.
* Remove tracing
* add noMessages, remove unnecessary getConnectionMessage from test
* add sending messages to the test
---------
Co-authored-by: IC Rainbow <aenor.realm@gmail.com>
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* server: add control port commands for clients and ghc threads (#835)
* Add stats-rts control query
With supporting ghc-options that would provide the data.
* Add CPSkip command
Allows spamming empty lines a few times to clean up the view.
* server: Add CP commands to enumerate clients and threads
* style
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* use base64 encoding for session ID
* fromMaybe
* whitespace
---------
Co-authored-by: Alexander Bondarenko <486682+dpwiz@users.noreply.github.com>
* Add env lookups for smp server paths
Allows running smp-server without touching system paths.
Can be helpful for running multiple instances.
* allow empty env var values
* diff
* fix
---------
Co-authored-by: Evgeny Poberezkin <2769109+epoberezkin@users.noreply.github.com>
* agent: catch IO errors correctly in MonadError
* correction
* correction
* utils
* agentFinally to catch IO exceptions in ExceptT
* rename
* remove, inline
* rename utils
* utils unit test
* test to show catch and finally problems
* tryAllErrors
* enable all tests
Add `XFTPClientAgent` — a per-server connection pool matching the Haskell pattern. The agent caches `XFTPClient` instances by server URL. All orchestration functions (`uploadFile`, `downloadFile`, `deleteFile`) take `agent` as first parameter and use `getXFTPServerClient(agent, server)` instead of calling `connectXFTP` directly. Connections stay open on success; the caller creates and closes the agent.
`connectXFTP` and `closeXFTP` stay exported (used by `XFTPWebTests.hs` Haskell tests). The `browserClients` hack, per-function `connections: Map`, and `getOrConnect` are deleted.
## Changes: client.ts
**Add** after types section: `XFTPClientAgent` interface, `newXFTPAgent`, `getXFTPServerClient`, `closeXFTPServerClient`, `closeXFTPAgent`.
**Delete**: `browserClients` Map and all `isNode` browser-cache checks in `connectXFTP` and `closeXFTP`.
**Revert `closeXFTP`** to unconditional `c.transport.close()` (browser transport.close() is already a no-op).
`connectXFTP` stays exported (backward compat) but becomes a raw low-level function — no caching.
## Changes: agent.ts
**Imports**: replace `connectXFTP`/`closeXFTP` with `getXFTPServerClient`/`closeXFTPAgent` etc.
**Re-export** from agent.ts: `newXFTPAgent`, `closeXFTPAgent`, `XFTPClientAgent`.
**`uploadFile`**: add `agent: XFTPClientAgent` as first param. Replace `connectXFTP` → `getXFTPServerClient`. Remove `finally { closeXFTP }`. Pass `agent` to `uploadRedirectDescription`.
**`uploadRedirectDescription`**: change from `(client, server, innerFd)` to `(agent, server, innerFd)`. Get client via `getXFTPServerClient`.
**`downloadFile`**: add `agent` param. Delete local `connections: Map`. Replace `getOrConnect` → `getXFTPServerClient`. Remove finally cleanup. Pass `agent` to `downloadWithRedirect`.
- allow receiving multiple messages from single iOS notification (#1355, #1362).
- prepare connection to accept to avoid race condition with events (#1365).
- transport isolation mode "Session" (default) to use new SOCKS credentials when client restarts or SOCKS proxy configuration changes (#1321).
Ntf server:
- control port (#1354).
- enable pings on ntf subscriptions, to resubscribe on reconnection (#1353).
SMP server:
- support multiple server ports (#1319).
- support serving HTTPS and SMP transport on the same port (#1326, #1327).
- persist iOS notifications to avoid losing them when Ntf server is offline (#1336, #1339, #1350).
- fix lost notification subscriptions (#1347).
- reject SKEY with different key earlier, at verification step (#1366).
- pass server information via CLI during server initialization (#1356).
- show version on server page (#1341).
- explicit graceful shutdown on SIGINT (#1360).
XRCP (remote access protocol):
- use SHA3-256 in hybrid key agreement (#1302).
- session encryption with forward secrecy (#1328).
# 6.0.5
SMP agent:
- support generic SOCKS proxy (without isolate-by-auth).
- reduce max message sizes
# 6.0.4
SMP server:
- better performance/memory: fewer map updates on re-subscriptions (#1297), split and reduce STM transactions (#1294)
- send DELD when subscribed queue is deleted (#1312)
- add created/updated/used date to queues to manage expiration (#1306)
XFTP server: truncate file creation time to 1 hour (#1310)
Servers:
- bind control port only to 127.0.0.1 for better security in case of firewall misconfiguration (#1280)
- reduce memory used for period stats (#1298)
Agent: process last notification from list (#1307)
- report receive file error with redirected file ID, when redirect is present (#1304)
- special error when deleted user record is not in database (#1303)
- fix race when sending a message to the deleted connection (#1296)
- support for multiple messages in a single notification
Ntf server:
- only use SOCKS proxy for servers without public address (#1314)
# 6.0.3
Agent:
- fix possible stuck queue rotation (#1290).
SMP server:
- batch END responses when subscribed client switches to reduce server and client traffic.
- reduce STM transactions for better performance.
- add stats for END events and for SUB/DEL event batches.
- remove "expensive" stats to save memory.
# 6.0.2
SMP agent:
- fix stuck connection commands when a server is not responding.
- store query errors, reduce slow query threshold to 1ms.
Notification server:
- reduce PING interval to 1 minute.
- fix subscriptions disabled on race condition (only mark subscriptions with END status when received via the active connection).
# 6.0.1
SMP agent:
- support changing user of the new connection.
- do not start delivery workers when there are no messages to deliver.
- enable notifications for all connections.
- combine database transactions when subscribing.
SMP server:
- safe compacting of store log.
- fix possible race when creating client that might lead to memory leak.
Dependencies: upgrade tls to 1.9
# 6.0.0
Version 6.0.0.8
Agent:
- enabled fast handshake support.
- batch-send multiple messages in each connection.
- resume subscriptions as soon as agent moves to foreground or as network connection resumes.
- "known" servers to determine whether to use SMP proxy.
- retry on SMP proxy NO_SESSION error.
- fixes to notification subscriptions.
- persistent server statistics.
- better concurrency.
SMP server:
- reduce threads usage.
- additional statistics.
- improve disabling inactive clients.
- additional control port commands for monitoring.
Notification server:
- support onion-only SMP servers.
# 5.8.2
Agent:
- fast handshake support (disabled).
- new statistics api.
SMP server:
- fast handshake support (SKEY command).
- minor changes to reduce memory usage.
# 5.8.1
Agent:
- API to reconnect one server.
- Better error handling of file errors and remote control connection errors.
- Only start uploading file once all chunks were registered on the servers.
SMP server:
- additional stats for sent message notifications.
- fix server page layout.
# 5.8.0
Version 5.8.0.10
SMP server and client:
- protocol extension to forward messages to the destination servers, to protect sending client IP address and transport session.
Agent:
- process timed out subscription responses to reduce the number of resubscriptions.
- avoid sending messages and commands when waiting for response timed out (except batched SUB and DEL commands).
- fix issue with stuck message reception on slow connection (when response to ACK timed out, and the new message was not processed until resubscribed).
- fix issue when temporary file sending or receiving error was treated as permanent.
SMP server:
- include OK responses to all batched SUB requests to reduce subscription timeouts.
XFTP server:
- report file upload timeout as TIMEOUT, to avoid delivery failure.
# 5.7.6
XFTP agent:
- treat XFTP handshake timeouts and network errors as temporary, to retry file operations.
# 5.7.5
SMP agent:
- fail if non-unique connection IDs are passed to sendMessages (to prevent client errors and deadlocks).
# 5.7.4
SMP agent:
- remove re-subscription timeouts (as they are tracked per operation, and could cause failed subscriptions).
- reconnect XFTP clients when network settings changes.
- fix lock contention resulting in stuck subscriptions on network change.
# 5.7.3
SMP/NTF protocol:
- add ALPN for handshake version negotiation, similar to XFTP (to preserve backwards compatibility with the old clients).
- upgrade clients to versions v7/v2 of the protocols.
SMP server:
- faster responses to subscription requests.
XFTP client:
- fix network exception during file download treated as permanent file error.
SMP agent:
- do not report subscription timeouts while client is offline.
# 5.7.2
SMP agent:
- fix connections failing when connecting via link due to race condition on slow network.
- remove concurrency limit when waiting for connection subscription.
- remove TLS timeout.
# 5.7.1
SMP agent:
- increase timeout for TLS connection via SOCKS
# 5.7.0
Version 5.7.0.4
_Please note_: the earliest SimpleX Chat clients supported by this version of the servers is 5.5.3 (released on February 11, 2024).
SMP server:
- increase max SMP protocol version to 7 (support for deniable authenticators).
NTF server:
- increase max NTF protocol version to 2 (support for deniable authenticators).
XFTP server:
- version handshake using ALPN.
SMP agent:
- increase timeouts for XFTP files.
- don't send commands after timeout.
- PQ encryption support.
# 5.6.2
Version 5.6.2.2.
SMP agent:
- Lower memory consumption (~20-25%).
- More stable XFTP file uploads and downloads.
- API to receive network connectivity changes from the apps.
- to reduce battery consumption: connection attempts interval growing to every 2 hours when app reports as offline.
- to reduce retries and traffic: 50% increased timeouts when on mobile network.
XFTP server:
- expire files on start.
- version negotiation based on TLS ALPN and handshake.
NTF server:
- reduced downtime by ~100x faster start time.
- exclude test tokens from statistics.
# 5.6.1
Version 5.6.1.0.
- Much faster iOS notification server start time (fewer skipped notifications).
- Fix SMP server stored message stats.
- Prevent overwriting uploaded XFTP files with subsequent upload attempts.
- Faster base64 encoding/parsing.
- Control port audit log and authentication.
# 5.6.0
Version 5.6.0.4.
SMP protocol/client/server:
- support deniable sender command authorization (to be enabled in the next version).
- remove support for SMP protocol versions (prior to v4, 07/2022).
Agent:
- optional post-quantum key agreement using sntrup761 in double ratchet protocol.
- improve performance of deleting multiple connections and files by batching database operations.
- delay connection deletion to deliver pending messages.
- API to test for notifications server.
- remove support for client protocols versions (prior to 10/2022).
XFTP server:
- restore storage quota in case of failed uploads.
Performance and stability improvements.
# 5.5.3
Agent:
- notification token API also returns active notifications server.
- support file descriptions with redirection and file URIs.
Servers:
- CLI commands for online key and certificate rotation.
- Configure config and log paths via environment variables.
# 5.5.2
Extensible handshake for clients and SMP/NTF servers (ignore extra data).
# 5.5.1
SMP servers:
- do not keep stats file open
- additional stats about currently stored messages
Agent:
- support multiple notification servers (only one can be used at a time).
- expire messages after "quota exceeded" error after 7 days (instead of 21 days previously).
- stabilize message delivery, remove unnecessary subscription retries and traffic.
- improve database performance for message delivery.
- fix sockets/memory leak - a very old bug "activated" by improvements in v5.5.0.
# 5.5.0
Code:
- compatible with GHC 8.10.7 to support compilation for armv7a.
- migrate to `crypton` from deprecated `cryptonite` (the seed for DRG is now sha512-hashed).
- use ChaChaDRG for all random IDs, keys and nonces, only using hashed entropy as seed.
- more efficient transaction batching in SMP protocol client and server.
Agent:
- stabilize message reception and delivery, migrate message delivery to database queue.
- additional event MSGNTF confirming that message received via notification is processed.
- efficient processing of messages sent to multiple recipients with batched database transactions.
- new worker abstraction for all queued tasks resilient to race conditions and some database errors.
- many fixed race conditions.
- background mode for iOS NSE.
- additional error reporting to client on critical errors (to be show as alert in the clients).
- functional api to get worker statistics.
SMP/XFTP servers:
- fix socket and memory leak on servers with high load (inactive clients without subscriptions are disconnected after set time of inactivity).
- control port improvements.
- fix statistics for stored queues, messages and files.
- make writing to store log atomic (fixes a rare bug in XFTP server).
- improve asynchronous connection deletion – it may now be completed after the client is restarted as well.
- improve subscription logic to retry if initial attempt fails.
- end SMP client connection after a number of failed PINGs (defalt is 3).
- end SMP client connection after a number of failed PINGs (default is 3).
# 4.3.0
@@ -105,7 +614,7 @@ SMP agent:
Notification server (v1.3.0):
- check token status when sending verification notificaiton.
- check token status when sending verification notification.
# 4.1.0
@@ -124,15 +633,15 @@ SMP agent:
SMP server:
- Basic authentication. The server address can now include an optional password that is required to create messaging queues, so the contacts who message you will not be able to receive messages via your server, unless you share with them the address with the password. It is recommended to enable basic authentication on all private servers by adding `create_password` parameter into AUTH section of server INI file (the previously deployed servers do not have this section, you need to add it).
- Disable creating new queues completely with `new_queues: off` parameter in AUTH section of INI file - it can be used to simplify migrating the exising connections to another server.
- Disable creating new queues completely with `new_queues: off` parameter in AUTH section of INI file - it can be used to simplify migrating the existing connections to another server.
- Updated server CLI with changed defaults:
- interactive server initialization, use -y flag to initalize the server non-interactively.
- interactive server initialization, use -y flag to initialize the server non-interactively.
- store log is now enabled by default, so messaging queues and messages are restored when the server is restarted (to restore undelivered messages the server needs to be stopped with SIGINT signal).
- a random password is now generated by default during the server initialization.
SMP agent:
- API to test SMP servers. It connects to the server, creates and deletes a messaging queue. The new SimpleX Chat client uses this API to allow you to test that you have the correct server address, with the valid certificate fingerprint and pasword, before enabling the new server.
- API to test SMP servers. It connects to the server, creates and deletes a messaging queue. The new SimpleX Chat client uses this API to allow you to test that you have the correct server address, with the valid certificate fingerprint and password, before enabling the new server.
# 3.4.0
@@ -178,9 +687,9 @@ SMP server and agent:
SMP server:
- restore undeliverd messages when the server is restarted.
- restore undelivered messages when the server is restarted.
- SMP protocol v3 to support push notification:
- updated SEND and MSG to add message flags (for notification flag that contros whether the notification is sent and for any future extensions) and to move message meta-data sent to the recipient into the encrypted envelope.
- updated SEND and MSG to add message flags (for notification flag that confirm whether the notification is sent and for any future extensions) and to move message meta-data sent to the recipient into the encrypted envelope.
- update NKEY and NID to add e2e encryption keys (for the notification meta-data encryption between SMP server and the client), and update NMSG to include this meta-data.
- update ACK command to include message ID (to avoid acknowledging unprocessed message).
- add NDEL commands to remove notification subscription credentials from SMP queue.
@@ -191,7 +700,7 @@ SMP agent:
- new protocol for duplex connection handshake reducing traffic and connection time.
- support for SMP notifications server and managing device token.
- remove redundant FQDN validation from TLS handshake to prepare for access via Tor.
- support for fully stopping agent and for termporary suspending agent operations.
- support for fully stopping agent and for temporary suspending agent operations.
- improve management of duplicate message delivery.
📢 SimpleXMQ v1 is released - with many security, privacy and efficiency improvements, new functionality - see [release notes](https://github.com/simplex-chat/simplexmq/releases/tag/v1.0.0).
@@ -116,7 +116,7 @@ On Linux, you can deploy smp and xftp server using Docker. This will download im
2. Run your Docker container.
- `smp-server`
You must change **your_ip_or_domain**. `-e "pass=password"` is optional variable to password-protect your `smp` server:
```sh
docker run -d \
@@ -129,7 +129,7 @@ On Linux, you can deploy smp and xftp server using Docker. This will download im
```
- `xftp-server`
You must change **your_ip_or_domain** and **maximum_storage**.
```sh
docker run -d \
@@ -149,8 +149,15 @@ On Linux, you can deploy smp and xftp server using Docker. This will download im
You can install and setup servers automatically using our script:
- Initialize SMP server with `smp-server init [-l] -n <fqdn>` or `smp-server init [-l] --ip <ip>` - depending on how you initialize it, either FQDN or IP will be used for server's address.
d="m 282.715,299.835 a 3.29,3.29 0 0 0 -2.662,5.336 l 9.474,12.261 A 7.894,7.894 0 0 0 289,320.257 v 18.21 a 7.877,7.877 0 0 0 7.895,7.895 h 84.21 A 7.877,7.877 0 0 0 389,338.468 v -18.211 c 0,-0.999 -0.19,-1.949 -0.525,-2.826 l 9.472,-12.26 a 3.29,3.29 0 0 0 -2.433,-5.334 3.29,3.29 0 0 0 -2.772,1.31 l -9.013,11.666 a 7.91,7.91 0 0 0 -2.624,-0.45 h -84.21 c -0.922,0 -1.8,0.163 -2.622,0.45 l -9.015,-11.666 a 3.29,3.29 0 0 0 -2.543,-1.312 z m 14.18,49.527 A 7.877,7.877 0 0 0 289,357.257 v 52.21 a 7.877,7.877 0 0 0 7.895,7.895 h 84.21 A 7.877,7.877 0 0 0 389,409.468 v -52.211 a 7.877,7.877 0 0 0 -7.895,-7.895 z"
<textx="-6"y="240"text-anchor="middle"font-family="system-ui, sans-serif"font-size="10"fill="#70F0F9"transform="rotate(-90 -6 240)">key in URL fragment - never sent to page server or data router</text>
<!-- Closed padlock: encryption (between sender and chunks) -->
<textx="-6"y="240"text-anchor="middle"font-family="system-ui, sans-serif"font-size="10"fill="#0053D0"transform="rotate(-90 -6 240)">key in URL fragment - never sent to page server or data router</text>
<!-- Closed padlock: encryption (between sender and chunks) -->
The implementation of sntrup761 is the _exact_ copy from this [Internet draft](https://www.ietf.org/archive/id/draft-josefsson-ntruprime-streamlined-00.html).
This file provides guidance on coding style and approaches and on building the code.
## Code Security
When designing code and planning implementations:
- Apply adversarial thinking, and consider what may happen if one of the communicating parties is malicious.
- Formulate an explicit threat model for each change - who can do which undesirable things and under which circumstances.
## Code Quality Standards
Haskell client and server code serves as system specification, not just implementation — we use type-driven design to reflect the business domain in types. Quality, conciseness, and clarity of Haskell code are critical.
## Code Style, Formatting and Approaches
The project uses **fourmolu** for Haskell code formatting. Configuration is in `fourmolu.yaml`.
**Key formatting rules:**
- 2-space indentation
- Trailing function arrows, commas, and import/export style
- Record brace without space: `{field = value}`
- Single newline between declarations
- Never use unicode symbols
- Inline `let` style with right-aligned `in`
**Format code before committing:**
```bash
# Format a single file
fourmolu -i src/Simplex/Messaging/Protocol.hs
```
Some files that use CPP language extension cannot be formatted as a whole, so individual code fragments need to be formatted.
**Follow existing code patterns:**
- Match the style of surrounding code
- Use qualified imports with short aliases (e.g., `import qualified Data.ByteString.Char8 as B`)
- Use record syntax for types with multiple fields
- Prefer explicit pattern matching over partial functions
**Comments policy:**
- Avoid redundant comments that restate what the code already says
- Only comment on non-obvious design decisions or tricky implementation details
- Function names and type signatures should be self-documenting
- Do not add comments like "wire format encoding" (Encoding class is always wire format) or "check if X" when the function name already says that
- Assume a competent Haskell reader
**Diff and refactoring:**
- Avoid unnecessary changes and code movements
- Never do refactoring unless it substantially reduces cost of solving the current problem, including the cost of refactoring
- Aim to minimize the code changes - do what is minimally required to solve users' problems
**Document and code structure:**
- **Never move existing code or sections around** - add new content at appropriate locations without reorganizing existing structure.
- When adding new sections to documents, continue the existing numbering scheme.
- Trace data flows end-to-end: from origin, through storage/parameters, to consumption. Flag values that are discarded and reconstructed from partial data (e.g. extracted from a URI missing original fields) — this is usually a bug.
- Read implementations of called functions, not just signatures — if duplication involves a called function, check whether decomposing it resolves the duplication.
- Do not save time on analysis. Read every function in the data flow even when the interface seems clear — wrong assumptions about internals are the main source of missed bugs.
### Haskell Extensions
- `StrictData` enabled by default
- Use STM for safe concurrency
- Assume concurrency in PostgreSQL queries
- Comprehensive warning flags with strict pattern matching
- **SMP Server**: Message broker with TLS, in-memory queues, optional persistence ([main code](../src/Simplex/Messaging/Server.hs), [all code files](../src/Simplex/Messaging/Server/), [executable](../apps/smp-server/)). For proxying SMP commands the server uses [lightweight SMP client](../src/Simplex/Messaging/Client/Agent.hs).
- **SMP Client**: Functional API with STM-based message delivery ([code](../src/Simplex/Messaging/Client.hs)).
- **SMP Agent**: High-level duplex connections via multiple simplex queues with E2E encryption ([code](../src/Simplex/Messaging/Agent.hs)). Implements Agent-to-agent protocol ([code](../src/Simplex/Messaging/Agent/Protocol.hs), [spec](../protocol/agent-protocol.md)) via intermediary agent client ([code](../src/Simplex/Messaging/Agent/Client.hs)).
- **XFTP**: SimpleX File Transfer Protocol, server and CLI client ([code](../src/Simplex/FileTransfer/), [spec](../protocol/xftp.md)).
- **XRCP**: SimpleX Remote Control Protocol ([code](`../src/Simplex/RemoteControl/`), [spec](../protocol/xrcp.md)).
- **Notifications**: Push notifications server requires PostgreSQL ([code](../src/Simplex/Messaging/Notifications), [executable](../apps/ntf-server/)). Client protocol is used for clients to communicate with the server ([code](../src/Simplex/Messaging/Notifications/Protocol.hs), [spec](../protocol/push-notifications.md)). For subscribing to SMP notifications the server uses [lightweight SMP client](../src/Simplex/Messaging/Client/Agent.hs).
## Architecture
For general overview see `../protocol/overview-tjr.md`.
We do not make code changes to improve code - any change must address a specific user problem or request.
## Discuss the plans as early as possible
Please discuss the problem you want to solve and your detailed implementation plan with the project team prior to contributing, to avoid wasted time and additional changes. Acceptance of your contribution depends on your willingness and ability to iterate the proposed contribution to achieve the required quality level, coding style, test coverage, and alignment with user requirements as they are understood by the project team.
## Follow project structure, coding style and approaches
./PROJECT.md has information about the structure of this `simplexmq` repository.
./CODE.md has details about general requirements common for `simplexmq` and `simplex-chat` repositories.
This files can be used with LLM prompts, e.g. if you use Claude Code you can create CLAUDE.md file in project root importing content from these files:
rts_heap grew from 10.1 GB to 20.7 GB over 5.5 hours.
Post-GC rts_live floor rose from 5.5 GB to 9.1 GB.
### Findings
**1. Large/pinned objects dominate live data (60-80%)**
`rts_large` = 4.5-8.0 GB out of 5.6-10.2 GB live. These are allocations > ~3KB that go on GHC's large object heap. They oscillate (not growing monotonically), meaning they are being allocated and freed constantly — transient, not leaked.
**2. Fragmentation is the heap growth mechanism**
`rts_heap ≈ rts_live + rts_frag`. The heap grows because pinned/large objects fragment GHC's block allocator. Once GHC expands the heap, it never shrinks. Growth pattern:
- Large objects allocated → occupy blocks
- Large objects freed → blocks can't be reused if ANY other object shares the block
- New allocations need fresh blocks → heap expands
- Heap never returns memory to OS
**3. Non-large heap data is stable (~1.0-2.2 GB)**
Normal Haskell objects (Maps, TVars, closures, client structures) account for only 1-2 GB. This scales with client count at ~100-130 KB/client and does NOT grow over time.
**4. All tracked data structures are NOT the cause**
- `clientSndQ=0, clientMsgQ=0` — TBQueues empty, no message accumulation
- `smpQSubs` oscillates ~1.0-1.4M — entries are cleaned up, not leaking
- `ntfStore` < 2K entries — negligible
- All proxy agent maps near 0
- `loadedQ=0` — useCache=False confirmed working
**5. Source of large objects is unclear without heap profiling**
The 4.5-8.0 GB of large objects could come from:
- PostgreSQL driver (`postgresql-simple`/`libpq`) — pinned ByteStrings for query results
- TLS library (`tls`) — pinned buffers per connection
- Network socket I/O — pinned ByteStrings for recv/send
- SMP protocol message blocks
Cannot distinguish between these without `-hT` heap profiling (which is too expensive for this server).
### Root Cause
**GHC heap fragmentation from constant churn of large/pinned ByteString allocations.**
Not a data structure leak. The live data itself is reasonable (5-10 GB for 15-17K clients). The problem is that GHC's copying GC cannot compact around pinned objects, so the heap grows with fragmentation and never shrinks.
### Mitigation Options
All are RTS flag changes — no rebuild needed, reversible by restart.
**1. `-F1.2`** (reduce GC trigger factor from default 2.0)
- Triggers major GC when heap reaches 1.2x live data instead of 2x
- Reclaims fragmented blocks sooner
- Trade-off: more frequent GC, slightly higher CPU
- Risk: low — just makes GC run more often
**2. Reduce `-A16m` to `-A4m`** (smaller nursery)
- More frequent minor GC → short-lived pinned objects freed faster
- Trade-off: more GC cycles, but each is smaller
- Risk: low — may actually improve latency by reducing GC pause times
**3. `+RTS -xn`** (nonmoving GC)
- Designed for pinned-heavy workloads — avoids copying entirely
- Available since GHC 8.10, improved in 9.x
- Trade-off: different GC characteristics, less battle-tested
- Risk: medium — different GC algorithm, should test first
Both `subscribers.queueSubscribers` and `ntfSubscribers.queueSubscribers` (and their `serviceSubscribers`) use `SubscribedClients (TMap EntityId (TVar (Maybe (Client s))))`.
Comment at line 376: *"The subscriptions that were made at any point are not removed"*
`deleteSubcribedClient` IS called on disconnect (Server.hs:1112) and DOES call `TM.delete`. But it only deletes if the current stored client matches — if another client already re-subscribed, the old client's disconnect won't remove the entry. This is by design for mobile client continuity, but the net effect on map size over time is unclear without measurement.
Comment at line 262: *"these vars are never removed, they are only added"*
These are only used for the proxy agent (SParty 'Sender), so they grow with each unique destination SMP server proxied to. With 292 unique servers in this log period, these are likely small — but long-running servers may accumulate thousands.
`closeSMPClientAgent` (line 369) does NOT clear these 4 maps.
#### 3. `NtfStore` — `NtfStore.hs:26`
`NtfStore (TMap NotifierId (TVar [MsgNtf]))` — one entry per NotifierId.
`deleteExpiredNtfs` (line 47) filters expired notifications from lists but does **not remove entries with empty lists** from the TMap. Over time, NotifierIds that no longer receive notifications leave zombie `TVar []` entries.
`deleteNtfs` (line 44) does remove the full entry via `TM.lookupDelete` — but only called when a notifier is explicitly deleted.
#### 4. `serviceLocks` in PostgresQueueStore — `Postgres.hs:112,469`
`serviceLocks :: TMap CertFingerprint Lock` — one Lock per unique certificate fingerprint.
`getCreateService` (line 469) calls `withLockMap (serviceLocks st) fp` which calls `getMapLock` (Agent/Client.hs:1029-1032) — this **unconditionally inserts** a Lock into the TMap. There is **no cleanup code** for serviceLocks anywhere. This is NOT guarded by `useCache`.
#### 5. `sentCommands` per proxy client connection — `Client.hs:580`
Each `PClient` has `sentCommands :: TMap CorrId (Request err msg)`. Entries are added per command sent (line 1369) and only removed when a response arrives (line 698). If a connection drops before all responses arrive, entries remain until the `PClient` is GC'd. Since `PClient` is captured by the connection thread which terminates on error, the `PClient` should become GC-eligible — but GC timing depends on heap pressure.
Unbounded `TQueue` for subscription changes. If the subscriber thread (`serverThread`) can't process changes fast enough, this queue grows without backpressure. With 22K clients subscribing/unsubscribing, sustained bursts could cause this queue to bloat.
10. **DB connection pool**: Bounded `TBQueue` with bracket-based return.
---
### Insufficient Data to Determine Root Cause
Without measuring the actual sizes of these structures at runtime, we cannot determine which (if any) is the primary contributor. The following exact logging changes will identify the root cause.
---
### EXACT LOGS TO ADD
Add a new periodic logging thread in `src/Simplex/Messaging/Server.hs`.
Insert at `Server.hs:197` (after `prometheusMetricsThread_`):
```haskell
<> memoryDiagThread_ cfg
```
Then define:
```haskell
memoryDiagThread_ :: ServerConfig s -> [M s ()]
memoryDiagThread_ ServerConfig {prometheusInterval = Just _} =
[memoryDiagThread]
memoryDiagThread_ _ = []
memoryDiagThread :: M s ()
memoryDiagThread = do
labelMyThread "memoryDiag"
Env { ntfStore = NtfStore ntfMap
, server = srv@Server {subscribers, ntfSubscribers}
, proxyAgent = ProxyAgent {smpAgent = pa}
, msgStore_ = ms
} <- ask
let interval = 300_000_000 -- 5 minutes
liftIO $ forever $ do
threadDelay interval
-- GHC RTS stats
rts <- getRTSStats
let liveBytes = gcdetails_live_bytes $ gc rts
heapSize = gcdetails_mem_in_use_bytes $ gc rts
gcCount = gcs rts
-- Server structures
clientCount <- IM.size <$> getServerClients srv
-- SubscribedClients (queue and service subscribers for both SMP and NTF)
Note: `smpSubs.subsCount` (queueSubscribers size) and `smpSubs.subServicesCount` (serviceSubscribers size) are **already logged** in Prometheus (lines 475-496). The log above adds all other candidate structures plus GHC RTS memory stats.
| `paWorkers` | Active reconnect workers | If growing, workers stuck in retry |
| `loadedQ` | Cached queues in store (0 with useCache=False) | Should be 0 |
| `ntfLocks` | Notifier locks in store | Should be 0 with useCache=False |
### Interpretation Guide
**If `smpQSubs` is in the millions**: SubscribedClients is the primary leak. Entries accumulate for every queue ever subscribed to.
**If `ntfStore` grows monotonically**: Zombie notification entries (empty lists after expiration). Fix: `deleteExpiredNtfs` should remove entries with empty lists.
**If `paActSvc` + `paActQ` grow**: Proxy agent subscription maps are the leak. Fix: add cleanup when no active/pending subs exist for a server.
**If `rts_live` is much smaller than `rts_heap`**: GHC heap fragmentation. Fix: tune `-F` flag (GC trigger factor) or use `-c` (compacting GC).
**If `rts_live` ~ 10-12GB**: The live data is genuinely large. Look at which metric is the largest contributor.
**If nothing above is large but `rts_live` is large**: The leak is in a structure not measured here — likely TLS connection buffers, ByteString retention from Postgres queries, or GHC runtime overhead. Next step would be heap profiling with `-hT`.
*)printf"${RED}Unsupported Ubuntu version!${NC}\nPlease file Github issue with request to support Ubuntu %s: https://github.com/simplex-chat/simplexmq/issues/new\n""$VERSION_ID"&&exit1;;
esac
version="$(printf'%s'"$VERSION_ID"| tr '.''_')"
arch="$(uname -p)"
case"$arch" in
x86_64)arch="$(printf'%s'"$arch"| tr '_''-')";;
*)printf"${RED}Unsupported architecture!${NC}\nPlease file Github issue with request to support %s architecture: https://github.com/simplex-chat/simplexmq/issues/new""$arch"&&exit1;;
Currently SMP routers store all queues in router memory. As the traffic grows, so does the number of undelivered messages. What is worse, Haskell is not avoiding heap fragmentation when messages are allocated and then de-allocated - undelivered messages use ByteString and GC cannot move them around, as they use pinned memory.
## Possible solutions
### Solution 1: solve only GC fragmentation problem
Move from ByteString to some other primitive to store messages in memory long term, e.g. ShortByteString, or manage allocation/de-allocation of stored messages manually in some other way.
Pros: the simplest solution that avoids substantial re-engineering of the router.
Cons:
- not a long term solution, as memory growth still has limits.
- may be ineffective, as it introduces additional copying of bytes.
### Solution 2: move message storage to hard drive
Use files or RocksDB to store messages.
Pros:
- much lower memory usage.
- no message loss in case of abnormal router termination (important until clients have delivery redundancy).
- this is a long term solution, and at some point it might need to be done anyway.
Cons:
- substantial re-engineering costs and risks.
- metadata privacy. Currently we only save undelivered messages when router is restarted, with this approach all messages will be stored for some time. this argument is limited, as hosting providers of VMs can make memory snapshots too, on the other hand they are harder to analyze than files. On another hand, with this approach messages will be stored for a shorter time.
#### RocksDB and other key-value stores
The downside of any key-value stores is that they don't seem to have efficient primitives for sequential delivery. While sequential delivery can be modelled with linked lists, they would require 1 key insert (on send), 3 key updates (1 update to update queue data on send, 1 update of the last message to point to the next, 1 update on delivery or message expiration) and 1 key deletion (on delivery or message expiration) for each delivered message.
This might result in substantial write amplification and compacting costs.
In general, tree structures that are efficient for quick lookups and updates, given approximately fixed value size, are inefficient for modelling queues.
#### Files
The upside of files is that they are well suited for sequential delivery and don't result in the same churn, with careful design, as trees do.
The downside of filesystem is that it does not scale well with the large number of files in a folder, so queues will have to be spread across multiple folders, following tree-like structure.
I could not find an available library that efficiently models sequential delivery in highly concurrent environment.
A possible design could be the following.
##### Queue folder and files
Each message queue is stored in its own folder (see below on folder locations). Folder would contain these files:
- `messages.abcd.log` - the file that is used to read messages from, sequentially
- `messages.efgh.log` - the optional file that is used to write messages, in case it is different from read file.
- `queue.log` - append-only file where the last line represents the current queue state
- `queue.timestamp.log` - previous states of queue.log file
Each line in "queue.log" file has this syntax
```abnf
queueLogLine =
%s"read_file=" base64
%s"read_msg=" digits
%s"read_byte=" digits
%s"write_file=" base64
%s"write_msg=" digits
```
When queue is first requested by the router:
```c
if queue folder exists:
read queue state from last line of queue.log
if queue.log contained more than one line: // compaction
copy queue.log to queue.timestamp.log
write one line queue state to queue.log
else:
create queue folder
create messages.abcd.log (abcd is some random string)
read_msg = 0
read_byte = 0
create queue.log with one line: "read_file=abcd read_msg=0 read_byte=0 write_files=abcd write_msg=0"
open read_file in ReadMode and seek to read_byte position
nextReadByte = read_byte
nextReadMsg = read_msg
open write_file in AppendMode
```
When message is added to the queue (assumes that queue state is loaded to router memory, if not the previous section will be done first):
```c
if write_msg > max_queue_messages:
return quota error
else if write_msg = max_queue_messages:
add quota_exceeded message to write_file
update queue state: write_msg += 1
append updated queue state to queue.log
else
// It is required that `max_queue_messages < max_file_messages`,
// so that we never need more than one additional write file.
if write_msg >= max_file_messages: // queue file rotation
create messages.efgh.log // efgh is some random string
update queue state: write_file=efgh write_msg=0 // read file remains the same as it was
append updated queue state to queue.log
copy queue.log to queue.timestamp.log
// `old` needs to be defined to limit the number and storage duration,
// preserving not more than N files, and not more than M days files, "and then some"
// (that is if the queue has high churn, we have file from M days before in any case, for any debugging).
delete `old``queue.timestamp.log` files
write one line queue state to queue.log // compaction
add message to write_file
update queue state: write_msg += 1
append updated queue state to queue.log
```
The above algorithm assumes `max_queue_messages < than max_file_messages`, so that we never need more than one write file.
When message is delivered, it is simply read from the read queue, queue state does not change yet:
```c
if nextReadMsg > read_msg:
deliver cached message, no need to read it again
else
read message from read_file handle
nextReadMsg = read_msg + 1
nextReadByte = current position in file
```
When message delivery is acknowledged, the read queue needs to be advanced, and possibly switched to read from the current write queue:
```c
if nextReadByte == read_byte:
return error // nothing was delivered
else if nextReadByte = EOF:
// end of file is reached, possibly some other condition,
// but it should allow changing max_file_messages on server restart
currReadFile = read_file
read_file = write_file
read_msg = 0
read_byte = 0
append updated queue state to queue.log
delete currReadFile
else
read_msg += 1
read_byte = nextReadByte
// `seek` should not be necessary, as the handle is already at nextReadByte position here
// seek to read_byte
append updated queue state to queue.log
```
The above algorithm delegates the problem of compaction and fragmentation management to file system, that is very optimized for such scenarios.
Also, read and write files will grow to almost a constant size, so the space they used may be re-used.
An important consideration is that writes to queue.log and message.log files and queue state modifications have to be sequential, without concurrency - it can be managed with the usual locks.
##### Queue folders structure
Most Linux systems use EXT4 filesystem where the file lookup time scales linearly to the number of files. While alternatives with logarithmic lookup time exist (XFS), they may be very complex to configure on the existing systems.
So storing all queue folders in one folder won't scale.
To solve this problem we could use recipient queue ID in base64url format not as a folder name, but as a folder path, splitting it to path fragments of some length. The number of fragments can be configurable and migration to a different fragment size can be supported as the number of queues on a given router grows.
Currently, queue ID is 24 bytes random number, thus allowing 2^192 possible queue IDs. If we assume that a router must hold 1b queues, it means that we have ~2^162 possible addresses for each existing queue. 24 bytes in base64 is 32 characters that can be split into say 8 fragments with 4 characters each, so that queue folder path for queue with ID `abcdefghijklmnopqrstuvwxyz012345` would be:
The maximum theoretic number of the folders on the 1st level is 64^4, or 2^24 ~ 16m - this is probably still a large number of subfolders for EXT4. Given that addresses are random, all the possible combinations in the first folder can be used with a large number of queues.
So we could use an unequal split of path, two letters each and the last being long:
The first three levels in this case can have 4096 subfolders each, and it gives 68b possible subfolders (64^2^3), so the last level will be sparse in case of 1b queues on the router. So we could make it 4 levels with 2 letters to never think about it, accounting for a large variance of the random numbers distribution:
Some networks block all ports other than web ports, including port 5223 used for SMP protocol by default. Running SMP routers on a common web port 443 would allow them to work on more networks. The routers would need to provide an HTTPS page for browsers (and probes).
## Problem
Browsers and tools rely on system CA bundles instead of certificate pinning.
The crypto parameters used by HTTPS are different from what the protocols use.
Public certificate providers like LetsEncrypt can only sign specific types of keys and Ed25519 isn't one of them.
This means a router should distinguish browser and protocol clients and adjust its behavior to match.
## Solution
`tls` package has a server hook that allows producing a different set of `TLS.Credentials` according to a client-provided "Server Name Indication" extension.
Since LE certificates are only handed out to domain names, TLS client will be sending the SNI.
However client transports are constructed over connected sockets and the SNI wouldn't be present unless explicitly requested.
When a client sends SNI, then it's a browser and web credentials should be used.
Otherwise it's a protocol client to be offered the self-signed ca, cert and key.
When a transport colocated with a HTTPS, its ALPN list should be extended with `h2 http/1.1`.
The browsers will send it, and it should be checked before running transport client.
If HTTP ALPN is detected, then the client connection is served with HTTP `Application` instead (the same "router information" page).
If some client connects to router IP, doesn't send SNI and doesn't send ALPN, it will look like a pre-handshake client.
In that case a router will send its handshake first.
This can be mitigated by delaying its handshake and letting the probe to issue its HTTP request.
## Implementation plan
An unmodified client should be able to use protocols on port 443 right away.
The switchover happens inside `runTransportServerState` before `runClient`:
```haskell
runServer (tcpPort, ATransport t) = do
-- ...
runTransportServerState_ ss started tcpPort serverParams tCfg $ \socket h -> do -- expose raw socket for warp-tls internals to attach
negotiated <- getSessionALPN
if allowHTTP t && isHTTP negotiated -- only attempt the switch for the TLS transport
then runHTTP socket (tlsContext h)-- ... collect data and produce values needed to run WAI Application
else runClient serverSignKey t h `runReaderT` env -- performs serverHandshake etc as usual
```
The web app and router live outside, so `runHttp` has to be provided by the `runSMPServer` caller.
Additonally, Warp is using its `InternalInfo` object that's scoped to `withII` bracket.
```haskell
runServer ini = do
-- ...
runWebServer ini ServerInformation {config, information} $ if sharedHttps then Nothing else webHttpsParams -- suppress serving https
if sharedHttps
then withRunHTTP staticFilesPath \attachStatic -> runSMPServer cfg (Just attachStatic) -- provide wrapped application runner
else runSMPServer cfg Nothing
```
### Upstream
The implementation relies on a few modification to upstream code:
- `warp-tls`: The library provides `httpOverTls`, but it wants to do handshake itself.
Since we have to do the handshake to switch on ALPN, the setup function has to be split.
This is a resonable change that may be upstreamed and nothing blocks us from using the recent version.
- `warp`: Only the re-export of `serveConnection` is needed.
Unfortunately the most recent `warp` version can't be used right away due to dependency cascade around `http-5` and `auto-update-2`.
So a fork containing the backported re-export has to be used until the dependencies are refreshed.
### TLS.ServerParams
When a router has port sharing enabled, a new set of TLS params is loaded and combined with transport params:
When `TRANSPORT.port` matches `WEB.https` the transport router becomes shared.
Perhaps a more desirable option would be explicit configuration resulting in additional transported to run:
```ini
[TRANSPORT]
port: 5223 ; pure protocol transport
# control_port: 5224
shared_port: 443 ; variant 1: register in TRANSPORT
[WEB]
https: 443
cert: /etc/opt/simplex/web.cert
key: /etc/opt/simplex/web.key
# transport: on ; variant 2:
```
## Caveats
Serving static files and the protocols together may pose a problem for those who currently use dedicated web servers as they should switch to embedded http handlers.
As before, using embedded HTTP server is increasing attack surface.
Users who want to run everything on a single host will have to add an extra IP address and bind routers to specific IPs instead of 0.0.0.0.
An amalgamated router binary can be provided that would contain both SMP and XFTP routers, where transport will dispatch connections by handshake ALPN.
## Alternative: Use transports routable with reverse-proxies
An "industrial" reverse proxy may do the ALPN routing, serving HTTP by itself and delegating `smp` and `xftp` to protocol servers.
Same with the `websockets`.
Since this in effect does TLS termination, the protocol routers will have to rely on credentials from protocol handshakes.
The journal storage routers recently migrated to do not delete delivered or expired messages, they only update pointers to journal file lines. The messages are actually deleted when the whole journal file is deleted (when fully deleted or fully expired).
The problem is that in case the queue stops receiving the new messages then writing of messages won't switch to the new journal file, and the current journal file containing delivered or expired messages would never be deleted.
## Solution
Remove current journal file and update queue_state.log during message expiration of "idle" queue (that is, without any new messages received or delivered within 3 hours) in case when:
- the queue is "empty" after the expiration
- the queue contains only quota marker(s), in which case move them to a new journal file and update the queue_state accordingly. Quota markers can be kept indefinitely to prevent writing the new messages to the dormant queues that reached capacity, so it's important to handle this case.
Also remove current journal file when the queue is opened in case it is empty (as it would not be ever expired in case it remains empty), and also update queue_state.log
Users report that message reception silently and permanently stops across all connections, with no error alerts. The app appears functional but no messages arrive. Recovery requires restart.
Root cause: a deadlock between worker threads holding `connLock` and the `agentSubscriber` (sole `subQ` reader).
### The deadlock mechanism
`subQ` (`TBQueue ATransmission`, capacity 4096 on mobile / 1024 on desktop) is the single pipeline between the agent layer and the chat layer. The `agentSubscriber` thread (`Commands.hs:4373`) is its **sole reader**.
Three code sites hold `connLock` and call blocking `writeTBQueue subQ` without a fullness check. When `subQ` is full, these block while holding the lock. If `agentSubscriber` simultaneously needs the same `connLock` (via `sendMessagesB_` → `withConnLocks`), it blocks too — creating a circular wait:
- **Worker**: holds `connLock(X)`, waits for `subQ` space (needs `agentSubscriber` to read)
- **agentSubscriber**: sole `subQ` reader, waits for `connLock(X)` (needs worker to release)
- **Result**: permanent silent deadlock — no exception, no alert, all connections blocked
### Confirmed deadlock scenarios
**Scenario 1**: Delivery worker during queue rotation test
No guard prevents a connection undergoing queue rotation (AM_QTEST_) or ACK processing from being included in `sendMessagesB_`'s batch. During these operations, the connection has `connStatus == ConnReady`, passing all filters in `memberSendAction`.
### Cascade amplification
Once any single deadlock triggers, `subQ` never drains. ALL other threads that attempt `writeTBQueue subQ` block progressively — their locks are held forever too. The entire threading system freezes within seconds.
2. **`nonBlockingWriteTBQueue`** (used at Client.hs:789, NtfSubSupervisor.hs:507):
```haskell
nonBlockingWriteTBQueue q x = do
sent <- atomically $ tryWriteTBQueue q x
unless sent $ void $ forkIO $ atomically $ writeTBQueue q x
```
Note: `nonBlockingWriteTBQueue` does NOT preserve ordering — the spawned background thread may complete out of order relative to subsequent direct writes from the same calling thread.
### Exhaustive proof: no other deadlock scenarios exist
All 15 `withConnLock` sites in Agent.hs were analyzed. Only 3 write to `subQ`:
| withConnLock site | Writes subQ? | Safe? |
|-------------------|-------------|-------|
| switchConnectionAsync' (899) | No | ✓ |
| setConnShortLinkAsync' (995) | No | ✓ |
| setConnShortLink' (1031) | No | ✓ |
| deleteConnShortLink' (1075) | No | ✓ |
| allowConnection' (1407) | No | ✓ |
| acceptContact' (1417) | No | ✓ |
| sendMessagesB_ (1708, `withConnLocks`) | No | ✓ |
Note: `processSMP` (line 3037) holds `connLock` and its local `notify` (line 3216) writes to `subQ`, but it uses the safe `isFullTBQueue` pattern. Its `ack` (line 3196) uses `enqueueCmd` (DB-only), NOT `ackQueueMessage`. The actual `ackQueueMessage` runs later from the async command worker via ICAck/ICAckDel.
Other lock pairs checked — no circular dependencies:
- `connLock × DB MVar`: DB never acquires connLock
- `entityLock × connLock`: consistent ordering (entity first in chat, conn in agent)
- `connLock(X) × connLock(Y)`: single agentSubscriber thread, one `withConnLocks` at a time
All deadlock paths require `agentSubscriber` to synchronously acquire `connLock`. Exhaustive analysis shows that **every such path converges on a single agent function**: `sendMessagesB_` → `withConnLocks` (Agent.hs:1708). No other agent API function called synchronously from the agentSubscriber acquires connLock.
Verified (FACT): `ackMessageAsync` → `enqueueCommand` only (no connLock). `toggleConnectionNtfs` → no lock. `deleteConnectionAsync` → `deleteLock` not `connLock`. `joinConnectionAsync` → `withInvLock` not `connLock`.
Also verified (FACT): `Lock = TMVar Text` (Lock.hs:24) is **non-reentrant** — double acquisition on the same thread deadlocks.
### All 22 trigger paths
Every path goes through `deliverMessage`/`deliverMessages`/`deliverMessagesB` → `withAgent sendMessagesB` → `sendMessagesB_` → `withConnLocks`:
1. **Single bottleneck**: All 22 paths converge on `sendMessagesB_` → `withConnLocks` (Agent.hs:1708). The deadlock is between this lock acquisition and any worker thread holding `connLock` + blocking on `writeTBQueue subQ`.
2. **Highest-risk paths** (#1, #2): Broadcasting to ALL group members in `introduceToAll` / `introduceToRemaining` acquires `withConnLocks` on ALL member connIds in a single batch. For large groups, this holds the agentSubscriber thread for a long time, during which subQ fills, which causes worker threads holding connLock on any of those connIds to deadlock.
3. **Medium-risk paths** (#5-7): `sendPendingGroupMessages` fires on every CON/SENT/QCONT. These are frequent and lock the member's connId, which is the SAME connId that a delivery worker or ACK worker may hold while writing to subQ.
---
## Analysis: `withConnLocks` in `sendMessagesB_`
### FACT: the lock protects ratchet encryption state
`sendMessagesB_` (Agent.hs:1708-1713) acquires `withConnLocks` and executes:
1. **`getConn_`** — reads connection metadata, send queues from DB
2. **`setConnPQSupport`** — updates PQ encryption flag per connection
3. **`enqueueMessagesB`** → `enqueueMessageB` → `storeSentMsg_` which calls:
- **`agentRatchetEncryptHeader`** (Agent.hs:3698) — reads current ratchet via `getRatchetForUpdate`, encrypts message header via `rcEncryptHeader`, writes advanced ratchet state via `updateRatchet`
- **`createSndMsg`** + **`createSndMsgDelivery`** — inserts message and delivery records
All operations run within `unsafeWithStore` → `withTransaction` (single DB transaction per batch).
### FACT: the lock CANNOT be removed
Without `withConnLocks`, concurrent `sendMessagesB_` calls targeting the same connection would:
- Read the same ratchet state, both encrypt, one overwrite the other → **ratchet desync** (unrecoverable)
- Get duplicate `internalSndId` values → **message ID collision**
- Race on `setConnPQSupport` → **PQ state inconsistency**
The lock serializes ALL operations on the connection's encryption state. Removing it would introduce data corruption.
Note: `sendMessage` (singular, line 530) uses the same `sendMessagesB_` function — there is no lock-free send path.
### Eliminated strategies
- **Strategy C (remove lock)**: The lock protects ratchet encryption. Removing it causes unrecoverable ratchet desync. Eliminated.
- **Strategy A (async dispatch)**: All 22 chat-layer callers use `deliverMessagesB` return values (delivery IDs, PQ state) synchronously. `forkIO` loses results. Eliminated.
- **Strategy W (isFullTBQueue + pending TVar)**: The existing pattern (lines 1937, 3216) buffers events in a local TVar and flushes after lock release. Between lock release and flush, another thread can acquire the same connLock and write events to subQ — reordering events within the same connection. This trades a visible deadlock for invisible ordering bugs. Eliminated.
- **Strategy O (per-connection overflow queues)**: Bounded overflow queues with "drop when full" were analyzed. Drop consequences are unacceptable at 5 of 6 write sites — CONF, INFO, CON cause permanent connection failure after ACK; INV loses connection invitations; SENT/MERR leave messages stuck forever. Unbounded overflow defeats backpressure. Eliminated.
---
## Solution: move subQ writes outside connLock
### Root cause
The `writeTBQueue subQ` calls at the 3 deadlock sites are inside `connLock` by accident of code structure, not necessity. `connLock` protects ratchet encryption state and DB consistency. The `notify` calls write informational events to `subQ` — they do not modify any state that `connLock` protects.
Moving the writes outside the lock scope eliminates the deadlock: blocking `writeTBQueue subQ` without holding `connLock` is safe — agentSubscriber is free to acquire the lock, process events, and drain `subQ`.
### Why reordering doesn't matter at these sites
The chat layer handlers for the 3 deadlock site events do NOT advance the ratchet:
Events that DO trigger ratchet advances (CON, SENT, QCONT → `sendPendingGroupMessages` → `sendMessagesB_`) are all already written OUTSIDE `connLock` in the current code.
Ratchet state lives in the DB, not in subQ events. agentSubscriber processes events sequentially regardless of arrival order. The SENT-before-SWITCH race already exists in the current code (new queue worker writes SENT outside connLock while old queue worker writes SWITCH inside connLock).
### Fix: Site 1 — `runSmpQueueMsgDelivery` AM_QTEST_ (line 2187)
Restructure `withConnLock` to return the event, write outside.
**Current code** (Agent.hs:2187-2214):
```haskell
AM_QTEST_ -> withConnLock c connId "runSmpQueueMsgDelivery AM_QTEST_" $ do
withStore' c $ \db -> setSndQueueStatus db sq Active
SomeConn _ conn <- withStore c (`getConn` connId)
case conn of
DuplexConnection cData' rqs sqs -> do
let addr = qAddress sq
case findQ addr sqs of
Just SndQueue {dbReplaceQueueId = Just replacedId, primary} ->
case removeQP (\sq' -> dbQId sq' == replacedId && not (sameQueue addr sq')) sqs of
Nothing -> internalErr msgId "sent QTEST: queue not found in connection"
Just (sq', sq'' : sqs') -> do
checkSQSwchStatus sq' SSSendingQTEST
atomically $ TM.delete (qAddress sq') $ smpDeliveryWorkers c
_ -> internalErr msgId "QTEST sent not in duplex ..." -- DEADLOCK
```
**New code:**
```haskell
AM_QTEST_ -> do
evt_ <- withConnLock c connId "runSmpQueueMsgDelivery AM_QTEST_" $ do
withStore' c $ \db -> setSndQueueStatus db sq Active
SomeConn _ conn <- withStore c (`getConn` connId)
case conn of
DuplexConnection cData' rqs sqs -> do
let addr = qAddress sq
case findQ addr sqs of
Just SndQueue {dbReplaceQueueId = Just replacedId, primary} ->
case removeQP (\sq' -> dbQId sq' == replacedId && not (sameQueue addr sq')) sqs of
Nothing -> pure $ Left "sent QTEST: queue not found in connection"
Just (sq', sq'' : sqs') -> do
checkSQSwchStatus sq' SSSendingQTEST
atomically $ TM.delete (qAddress sq') $ smpDeliveryWorkers c
withStore' c $ \db -> do
when primary $ setSndQueuePrimary db connId sq
deletePendingMsgs db connId sq'
deleteConnSndQueue db connId sq'
let sqs'' = sq'' :| sqs'
conn' = DuplexConnection cData' rqs sqs''
cStats <- connectionStats c conn'
pure $ Right $ SWITCH QDSnd SPCompleted cStats
_ -> pure $ Left "sent QTEST: there is only one queue in connection"
_ -> pure $ Left "sent QTEST: queue not in connection or not replacing another queue"
_ -> pure $ Left "QTEST sent not in duplex connection"
-- subQ write is now OUTSIDE connLock — blocking writeTBQueue is safe
case evt_ of
Right evt -> notify evt
Left err -> internalErr msgId err
```
All DB operations remain inside the lock. Only `notify`/`internalErr` (which write to subQ) move outside. `internalErr` calls `notifyDel` = `notify >> delMsg` — both `notify` (subQ write) and `delMsg` (`deleteSndMsgDelivery`, keyed on unique msgId) are safe outside the lock. The existing double-delete pattern (`delMsg` inside `internalErr` + `delMsgKeep` at line 2216) is preserved.
**Caller 2: `ICAck` / `ICAckDel`** (Agent.hs:1823-1824) — inline `tryWithLock` as `tryCommand` + `withConnLock`, write subQ between the two scopes:
`tryWithLock name = tryCommand . withConnLock c connId name` — by inlining, the subQ write can be placed outside `withConnLock` but inside `tryCommand` (retaining retry/error handling).
-- subQ write is OUTSIDE connLock — cannot deadlock with agentSubscriber
forM_ t_ $ atomically . writeTBQueue subQ
```
Where `ack` now returns `AM (Maybe ATransmission)`:
```haskell
ack srv rId srvMsgId = do
rq <- withStore c $ \db -> getRcvQueue db connId srv rId
ackQueueMessage c rq srvMsgId
```
All subQ writes for MSGNTF are now outside connLock. FIFO ordering is preserved — no `nonBlockingWriteTBQueue`, no forked threads. The same thread that held the lock writes to subQ sequentially after releasing it.
### Race analysis
Window between connLock release and subQ write at Site 1:
note over B: having received file description<br>from sender
loop for each chunk
B ->> S: 1a. download chunk ("FGET")
S ->> B: send chunk body ("FILE")
opt
B ->> S: 1b. acknowledge chunk reception ("FACK")
note over S: delete recipient ID
S ->> B: respond with ok ("OK")
end
end
note over B: 2. combine chunks into a file<br>3. decrypt file using key from file description<br>4. extract file name and unpad the file<br>5. validate file digest with the file description
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.