Conversation
The more important issue is that we should have drained the internal `respMux` subscription after all user subscriptions have been drained so that we don't miss replies that would be sent by the user subscriptions callbacks. Draining the internal `respMux` subscription would often cause the check for number of subscriptions to be positive, which then would lead to a wait of 100ms before checking again that the number would be down to 0. This PR addresses both issue: make sure that the internal subscription is drained last, and use condition variable to notify when subscriptions are removed and the number is down to the expected number. Resolves #911 Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #915 +/- ##
=======================================
Coverage 70.21% 70.21%
=======================================
Files 48 48
Lines 17271 17316 +45
Branches 3545 3555 +10
=======================================
+ Hits 12127 12159 +32
- Misses 1735 1738 +3
- Partials 3409 3419 +10 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
mtmk
approved these changes
Sep 19, 2025
| s = natsConn_enqueueUnsubProto(nc, sub->sid); | ||
| sid = sub->sid; | ||
| natsSub_Unlock(sub); | ||
| s = natsConn_enqueueUnsubProto(nc, sid); |
Member
There was a problem hiding this comment.
observation: are we taking the risk of unsubbing multiple times since it would be a noop on the server side?
Member
Author
There was a problem hiding this comment.
What makes you think we would be UNSUB'ing multiple times? The change here is probably not necessary, but felt like I did not want to have the sub's lock while in the function the enqueue the protocol. The caller has both the connection lock and subMu lock, so there is no risk that this subscription is UNSUB'ed twice here.
Member
There was a problem hiding this comment.
makes sense. I was only looking at this change that the natsConn_enqueueUnsubProto is out side the lock now.
kozlovic
added a commit
that referenced
this pull request
Sep 26, 2025
Release notes will be: This release contains some breaking changes. See the "Changed" section below. * Build * Disable NATS Streaming by default by @mtmk in #770 * TLS * Require OpenSSL 1.1.1+ to compile. Removed the `NATS_BUILD_TLS_USE_OPENSSL_1_1_API` CMake variable by @kozlovic in #905 * The option `natsOptions_SetSSLVerificationCallback` signature was changed to replace the use of `SSL_verify_cb` (which required OpenSSL dependency in the `nats.h` file), to the new callback `natsSSLVerifyCb`. See documentation of `natsSSLVerifyCb` to see the cast needed to compile with this new header file by @kozlovic in #908 * Modification of a `natsOptions` object if it has TLS/SSL configuration and is actively used by connections will now return a `NATS_ILLEGAL_STATE` by @kozlovic in #912 * Options * Ability to load the trusted CA certificates from a directory using the new option `natsOptions_LoadCATrustedCertificatesPath` by @kerbert101 in #862 * Ability to connect via HTTP proxy for instance by adding a proxy connection handler using the new option `natsOptions_SetProxyConnHandler` by @wolfkor in #871 and @kozlovic in #897 * Ability to load the certificate chain and key from a file on every connection attempt using the new option `natsOptions_LoadCertificatesChainDynamic` by @Matus-p in #901 * Ability to perform concurrent TLS handshakes that may improve time it takes for concurrent connections to be established using the new option `natsOptions_AllowConcurrentTLSHandshakes ` by @kozlovic in #914. Issue was reported by @yanyongcheng in #899 * JetStream * Per-message TTL support (a NATS Server v2.11 feature) by @levb in #863 * Pull consumer priority groups (a NATS Server v2.11 feature) by @levb in #869 * ObjectStore support by @kozlovic in #902. Thanks to @jfflynn41 and @alex1891 for the feedback in #876 * JetStream * Handling of publish asynchronous timeouts by @kozlovic in #886. Issue reported by @yanyongcheng in #880 * Timer insertion by @kozlovic in #883. Issue reported by @yanyongcheng in #881 * EventLoop: * Handling of possible failure on initial attach by @kozlovic in #918 * LibEvent: `natsConnection_Close()` not closing the TCP connection by @kozlovic in #882. Issue was reported by @yanyongcheng in #879 * Libuv: Possible crash if connection is destroyed while receiving data by @kozlovic in #889. Issue was reported by @yanyongcheng in #888 * KeyValue * Keys, History or watcher's next may incorrectly return `NATS_TIMEOUT` by @kozlovic in #917/ Issue was reported by @ArashPartow in #916 * MicroServices: * Wrong marshaling of `average_processing_time` by @kozlovic in #892. Issue was reported by @Archie3d in #890 * Statistics error was always incremented by @kozlovic in #894. Issue was reported by @Archie3d in #893 * TLS * Unknown type name `SSL_verify_cb` by @kozlovic in #878. Issue was reported by @philipfoulkes in #877 * Initialization and cleanup code related to OpenSSL was removed since it was deprecated for versions post OpenSSL 1.1. A cleanup function pertinent to 1.1+ code was possibly causing a problem. By @kozlovic in #905. Issue was reported by @vdeters in #904 * Possible hang during handshake by @kozlovic in #907. Issue was reported by @etrochim in #906 * Protect calls to `SSL_read` and `SSL_write` with a mutex. Since the same `SSL` object is shared between different threads, the OpenSSL library requires a mutex to be used by @kozlovic in #913 * Memory allocation check by @wooffie in #868 * Add missing status text string by @oldnick85 in #872 and @kozlovic in #874 (the issue was not present in any published release and was introduced in #869) * Parsing of message headers with `NULL` or all-whitespace values by @habbbe in #873 * Removed some unused code related to handling of responses and added custom inbox with very long prefix test by @kozlovic in #885. Issue was reported by @yanyongcheng in #884 * Connection drain could cause missed reply and/or a 100ms delay by @kozlovic in #915. Issue was reported by @T-Maxxx in #911 * Build * Deprecated Ubuntu 20.04 in GitHub actions by @levb in #864 * Removed the older compiler jobs by @levb in #865 * Fixed Windows build to use the NATS Server main branch by @levb in #866 * @kerbert101 made their first contribution in #862 * @habbbe made their first contribution in #873 * @wolfkor made their first contribution in #871 * @Matus-p made their first contribution in #901 Signed-off-by: Ivan Kozlovic <ivan@synadia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The more important issue is that we should have drained the internal
respMuxsubscription after all user subscriptions have been drained so that we don't miss replies that would be sent by the user subscriptions callbacks.Draining the internal
respMuxsubscription would often cause the check for number of subscriptions to be positive, which then would lead to a wait of 100ms before checking again that the number would be down to 0.This PR addresses both issue: make sure that the internal subscription is drained last, and use condition variable to notify when subscriptions are removed and the number is down to the expected number.
Resolves #911
Signed-off-by: Ivan Kozlovic ivan@synadia.com