>From my perspective, disabling the pooling (like the PR does) OR having the ability to set a time to live (TTL) for the pooled connections, should be good enough for tackling the current problem.
Michael Am Mo., 27. März 2023 um 12:41 Uhr schrieb Tamás Cservenák < ta...@cservenak.net>: > Created https://github.com/apache/maven-resolver/pull/274 > > T > > On Mon, Mar 27, 2023 at 11:04 AM Tamás Cservenák <ta...@cservenak.net> > wrote: > > > In this case, is "configuration of pool" (I guess you mean like sizes > etc) > > really needed? > > Would "use pool" (on/off) enough instead? > > > > T > > > > On Mon, Mar 27, 2023 at 10:54 AM Michael Vitz > > <vitz.mich...@googlemail.com.invalid> wrote: > > > >> Yes, it seems the isStale-Implementation of BHttpConnectionBase > >> > >> ``` > >> @Override > >> public boolean isStale() throws IOException { > >> if (!isOpen()) { > >> return true; > >> } > >> try { > >> final int bytesRead = > >> fillInputBuffer(Timeout.ofMilliseconds(1)); > >> return bytesRead < 0; > >> } catch (final SocketTimeoutException ex) { > >> return false; > >> } catch (final SocketException ex) { > >> return true; > >> } > >> } > >> ``` > >> > >> does not work correctly with the way Azure intercepts these connections. > >> > >> What's the way to improve the situation? Opening a bug there and waiting > >> for it to be fixed? Or is the option to allow configuration of the pool, > >> TTL and or Disabling the pool, like wagon did feasible? > >> > >> Am Mo., 27. März 2023 um 09:58 Uhr schrieb Konrad Windszus < > >> konra...@gmx.de > >> >: > >> > >> > Thanks for the reproducer and sorry for not reading thoroughly before. > >> > Indeed the connection pool is not configurable yet with the native > >> > connector. According to > >> > > >> > https://hc.apache.org/httpcomponents-client-4.5.x/current/httpclient/apidocs/org/apache/http/impl/conn/PoolingHttpClientConnectionManager.html > >> > stale connections are checked after 2 seconds before being reused but > >> > probably the check does not work for some reason with Azure…. > >> > > >> > > Am 27.03.2023 um 09:49 schrieb Michael Vitz < > >> vitz.mich...@googlemail.com > >> > .invalid>: > >> > > > >> > > Hi Konrad, > >> > > > >> > > The only two timeout related properties are > >> > aether.connector.requestTimeout > >> > > and aether.connector.connectTimeout. > >> > > Both do not solve my problem. > >> > > > >> > > I created a GitHub project ( > >> > > https://github.com/mvitz/maven-39-gh-actions-timeouts) which > >> reproduces > >> > the > >> > > problem. > >> > > It declares many dependencies/plugins to make sure the complete HTTP > >> pool > >> > > is used before the tests, contains a test that runs in total about 6 > >> > > minutes to take longer than the four-minute Azure timeout and uses > >> Maven > >> > > 3.9.1. > >> > > > >> > > The first build ( > >> > > > >> > > >> > https://github.com/mvitz/maven-39-gh-actions-timeouts/actions/runs/4524745997/jobs/7968775993 > >> > ) > >> > > uses -Daether.connector.requestTimeout=270000 which is longer than > >> four > >> > > minutes and fails. > >> > > The second one ( > >> > > > >> > > >> > https://github.com/mvitz/maven-39-gh-actions-timeouts/actions/runs/4524925829/jobs/7969075725 > >> > ) > >> > > uses -Daether.connector.connectTimeout=60000 > >> > > -Daether.connector.requestTimeout=180000 which is shorter but fails, > >> too. > >> > > > >> > > Both builds fail with: > >> > > Error: Failed to execute goal > >> > > org.apache.maven.plugins:maven-jar-plugin:3.3.0:jar (default-jar) on > >> > > project maven-39-gh-actions-timeouts: Execution default-jar of goal > >> > > org.apache.maven.plugins:maven-jar-plugin:3.3.0:jar failed: Plugin > >> > > org.apache.maven.plugins:maven-jar-plugin:3.3.0 or one of its > >> > dependencies > >> > > could not be resolved: Failed to collect dependencies at > >> > > org.apache.maven.plugins:maven-jar-plugin:jar:3.3.0 -> > >> > > org.apache.maven.shared:file-management:jar:3.1.0: Failed to read > >> > artifact > >> > > descriptor for org.apache.maven.shared:file-management:jar:3.1.0: > The > >> > > following artifacts could not be resolved: > >> > > org.apache.maven.shared:file-management:pom:3.1.0 (absent): Could > not > >> > > transfer artifact org.apache.maven.shared:file-management:pom:3.1.0 > >> > from/to > >> > > central (https://repo.maven.apache.org/maven2): Read timed out -> > >> [Help > >> > 1] > >> > > > >> > > I suspect that the HTTP pool is fully used before the tests, all > >> > > connections are silently interrupted during the test run that takes > >> > longer > >> > > than the Azure limit of four minutes, and the pool unable to recover > >> any > >> > > connection from that. > >> > > > >> > > As told before when using wagon with > -Dmaven.resolver.transport=wagon > >> and > >> > > configuring a pool time to love of 3 minutes via > >> > > -Dmaven.wagon.httpconnectionManager.ttlSeconds=180 everything works > as > >> > > expected ( > >> > > > >> > > >> > https://github.com/mvitz/maven-39-gh-actions-timeouts/actions/runs/4529868249/jobs/7978093099 > >> > > ). > >> > > > >> > > Regards, > >> > > Michael > >> > > > >> > > > >> > >> Am So., 26. März 2023 um 12:54 Uhr schrieb Konrad Windszus < > >> > konra...@gmx.de > >> > >>> : > >> > >> > >> > >> The timeouts are configurable even with new native connector: > >> > >> https://maven.apache.org/resolver/configuration.html. > >> > >> Please try if those settings help. > >> > >> Konrad > >> > >> > >> > >>> Am 25.03.2023 um 15:08 schrieb Michael Vitz < > >> > vitz.mich...@googlemail.com > >> > >> .invalid>: > >> > >>> > >> > >>> Hi all, > >> > >>> > >> > >>> We recently switched from Maven 3.8.x to 3.9.x and all of a > sudden, > >> we > >> > >> ran > >> > >>> into connection timeouts during downloading the maven-jar-plugin > >> after > >> > >> all > >> > >>> tests passed. > >> > >>> > >> > >>> After some digging, I suspect that it is a combination of GitHub > >> > Actions > >> > >>> running on Azure which silently drops open connections after > around > >> > four > >> > >>> minutes (see discussion in > >> > >>> https://github.com/actions/runner-images/issues/1499) and the > >> switch > >> > to > >> > >> the > >> > >>> native HTTP transport in Maven 3.9.x. > >> > >>> Our tests take quite some time and because the maven-jar-plugin is > >> > >> freshly > >> > >>> downloaded after these, the used connection was opened before the > >> tests > >> > >>> were started and is dropped by Azure meanwhile. > >> > >>> > >> > >>> Our current workaround is switching to the old Wagon HTTP provider > >> > (with > >> > >>> -Dmaven.resolver.transport=wagon) and setting a TTL for the used > >> HTTP > >> > >> pool > >> > >>> (-Dmaven.wagon.httpconnectionManager.ttlSeconds=180) or disabling > >> the > >> > >> pool > >> > >>> and keep alive (-Dhttp.keepAlive=false > >> -Dmaven.wagon.http.pool=false). > >> > >>> > >> > >>> Unfortunately, the new HTTP transport does not allow changing the > >> TTL > >> > >>> (which by default is -1 which means forever) or disabling it > >> > altogether. > >> > >> It > >> > >>> would be nice if such settings would be added in one of the next > >> > >> releases. > >> > >>> > >> > >>> I would be happy to help/try to provide a patch on my own if this > >> > helps. > >> > >>> > >> > >>> Regards, > >> > >>> Michael > >> > >> > >> > > >> > > >
