fix #289 Register pool-releasing close listener only if acquired #290
Conversation
…tually acquired
only set the channel closeFuture to release to the pool once we've made sure that we acquired a Channel.
Codecov Report
@@ Coverage Diff @@
## master #290 +/- ##
============================================
+ Coverage 67.78% 68.46% +0.68%
- Complexity 1015 1026 +11
============================================
Files 73 73
Lines 4287 4291 +4
Branches 611 613 +2
============================================
+ Hits 2906 2938 +32
+ Misses 1011 989 -22
+ Partials 370 364 -6
Continue to review full report at Codecov.
|
|
please heavily challenge the relevance of this fix. adding a little background on the issue itself. |
| public void fixedPoolTwoAcquire() | ||
| throws ExecutionException, InterruptedException, IOException { | ||
| final ScheduledExecutorService service = Executors.newScheduledThreadPool(2); | ||
| int echoServerPort = SocketUtils.findAvailableTcpPort(); |
There was a problem hiding this comment.
FYI, this is not entirely reliable and can cause problems on CI servers, if some other process grabs the socket. We've seen this happen once in a while, causing bogus CI build failures.
In my projects, where possible, we've changed tests to listen on port 0 (so the OS selects the port) and then query the test server for which port was actually selected.
Not critical but worth consideration.
There was a problem hiding this comment.
hard to do as the first part of the test checks what happens when the host is initially unreachable. crossing fingers for CI :(
* Fix `StompServerIntegrationTests` for disabling connection pool to workaround the regression in the Reactor-Netty-0.7.4 Related to reactor/reactor-netty#290
* Fix `StompServerIntegrationTests` for disabling connection pool to workaround the regression in the Reactor-Netty-0.7.4 Related to reactor/reactor-netty#290
Instead of setting up the close listener in the ChannelPoolHandler
channelCreated method, set it up in the PooledClientContextHandler.
This should prevent the listener to be invoked after a FixedChannelPool
has determined the channel was not healthy and has already decremented
its internal acquire counter (which would lead to a second decrement of
said counter and an assertion error).