summaryrefslogtreecommitdiffstats
path: root/messagebus/src
Commit message (Collapse)AuthorAgeFilesLines
* Make SourceSession pending count atomic to avoid data racesTor Brede Vekterli2022-03-142-20/+25
| | | | | Could have used the existing mutex, but this value is polled frequently by visitor threads so avoiding having to take a lock every time makes sense.
* avoid race in trace messages regarding pending countHåvard Pettersen2022-03-141-3/+6
|
* Remove inlining warnings (messagebus).Tor Egge2022-02-266-0/+32
|
* Use a typesafe Duration instead of integer milliseondst status.Henning Baldersheim2022-02-252-2/+5
|
* - Use a utility method that recomputes a timeout if VESPA_TIMER_HZ deviates ↵Henning Baldersheim2022-02-252-2/+2
| | | | from the default of 1000hz.
* Merge branch 'master' into balder/consider_vespa_timer_hzHenning Baldersheim2022-02-251-3/+9
|\
| * - Only wait with timeout if you have any recurring tasks to execute.Henning Baldersheim2022-02-252-4/+9
| | | | | | | | | | - Increase timeout from 10 -> 100 milliseconds. - Also consider the VESPA_TIMER_HZ control of timer frequency.
* | Take VESPA_TIMER_HZ into consideration.Henning Baldersheim2022-02-251-2/+2
|/
* Revert "Revert "- Create the common transport and threadpool in the main loop.""Henning Baldersheim2022-02-241-2/+2
|
* Revert "- Create the common transport and threadpool in the main loop."Henning Baldersheim2022-02-241-2/+2
|
* - Create the common transport and threadpool in the main loop.Henning Baldersheim2022-02-241-2/+2
| | | | | - Also use the common transport for config subscriptions. - Put The TransportConfig in the fnet namespace.
* Merge pull request #21285 from vespa-engine/vekterli/more-threading-fixesHenning Baldersheim2022-02-214-18/+34
|\ | | | | More miscellaneous threading fixes [run-systemtest]
| * Let `erase()` return new iterator instead of doing it manuallyTor Brede Vekterli2022-02-211-1/+1
| |
| * Use Kill() instead of Unschedule() to prevent against unwanted re-schedulingTor Brede Vekterli2022-02-211-1/+1
| |
| * Avoid potential MessageBus<->FNET mutex order inversionTor Brede Vekterli2022-02-182-7/+16
| | | | | | | | | | | | | | | | Destruct targets outside lock to prevent the following potential deadlock: 1. flushTargets (pool lock) -> FNET transport thread event (transport thread lock) 2. FNET CheckTasks (transport thread lock) -> periodic flushTargets task run -> flushTargets (pool lock) Also explicitly unschedule any target pool task on network shutdown
| * Make MessageBus resend-queue thread safeTor Brede Vekterli2022-02-182-11/+18
| | | | | | | | | | May be accessed from both sender thread (in case of synchronous send failure) as well as periodic resending from MBus thread.
* | Make ConfigUri constructors explicit and use same context where possible in ↵Henning Baldersheim2022-02-201-1/+1
|/ | | | proton.
* Reduce code visibility and include only what you need from config library.Henning Baldersheim2022-02-065-5/+2
|
* Run dynamic throttling tests fewer times by defaultJon Marius Venstad2022-02-031-7/+7
|
* Add SuppressWarnings for removal of config.subscription types.gjoranv2022-02-023-0/+3
|
* Sync with messagebus thread when disconnectingBjørn Christian Seime2022-01-252-2/+2
|
* Add disconnect() that unregisters sessionBjørn Christian Seime2022-01-253-0/+5
|
* TransportThread must always do a handover. If not you might get a deadlock.Henning Baldersheim2022-01-251-1/+1
|
* We must ensure that we do not end up with a rejected execution that can ↵Henning Baldersheim2022-01-241-3/+3
| | | | | | cause deadlock. Using an unbound Q here is fine as there is external throttling in place.
* Don't retry after resender is destroyedBjørn Christian Seime2022-01-211-10/+15
|
* Update C++ DynamicThrottlePolicy to match Java impl semanticsTor Brede Vekterli2022-01-173-37/+68
| | | | | | | | | | | This is an attempt at making the C++ implementation 1-1 with the Java implementation, the latter considered the source of truth for this matter. Unit tests have been updated to match their Java equivalents as well. This task was approached much in the same way a novice wizard would transcribe arcane runes that may possible blow their hands off; carefully and with more pragmatic respect than understanding.
* Use a daemon thread have wait for shutdown to complete.Henning Baldersheim2022-01-071-1/+1
|
* decouple tests by giving them separate destination session namesArne H Juul2021-12-301-64/+71
|
* speed up unit testArne H Juul2021-12-171-11/+132
| | | | | | | * use shared (static) slobrok, source and destination servers * reset various state between tests * still create/destroy sessions between tests * add logging of start/finish of each sub-test to track where time goes
* more aggressive waitStateArne H Juul2021-12-171-3/+9
|
* most unit tests can run with just 1 network threadArne H Juul2021-12-162-3/+5
|
* improve testserver start/stop logging slightlyArne H Juul2021-12-162-5/+16
|
* wait a bit more before spamming the logArne H Juul2021-12-151-2/+3
|
* improve signal-to-noise ratio in informational loggingArne H Juul2021-12-151-4/+10
|
* log some info to make it possible to see what is going on in unit testsArne H Juul2021-12-151-0/+5
|
* Remove the need for SyncableHenning Baldersheim2021-11-261-2/+1
|
* add some information while waiting for messagebus networkArne H Juul2021-11-112-1/+15
|
* Add missiung copyrightsJon Bratseth2021-10-202-0/+2
|
* fix unstable unit testHåvard Pettersen2021-10-151-0/+3
|
* Update Verizon Media copyright notices.gjoranv2021-10-074-4/+4
|
* Update 2017 copyright notices.gjoranv2021-10-07379-385/+385
|
* Merge pull request #19285 from ↵Jon Marius Venstad2021-09-272-16/+25
|\ | | | | | | | | vespa-engine/jonmv/shut-down-RPC-when-no-message-bus-owner-remain Shut down network when all owners also have detached
| * Better namesJon Marius Venstad2021-09-272-12/+14
| |
| * Shut down network when all owners also have detachedJon Marius Venstad2021-09-242-14/+21
| |
* | Simplify with computeIfAbsentHenning Baldersheim2021-09-251-7/+1
| |
* | Avoid using a thread local map as it keeps large objects hanging around as ↵Henning Baldersheim2021-09-241-15/+18
| | | | | | | | ghosts.
* | As a simple tcp based rpc address is the most common, split into Tcp and Named.Henning Baldersheim2021-09-246-56/+88
|/ | | | Tcp variant is immutable and hence thread safe, while Named is made thread safe by synchonized.
* Revert "Revert "Avoid copying data just to compress them when it is not ↵Henning Baldersheim2021-08-301-4/+2
| | | | necessary.""
* Revert "Avoid copying data just to compress them when it is not necessary."Henning Baldersheim2021-08-301-2/+4
|
* Avoid copying data just to compress them when it is not necessary.Henning Baldersheim2021-08-301-4/+2
|