You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@ignite.apache.org by "Ivan Daschinskiy (Jira)" <ji...@apache.org> on 2020/10/09 08:55:00 UTC
[jira] [Created] (IGNITE-13564) Improve SYSTEM_WORKER_BLOCKED
reporting.
Ivan Daschinskiy created IGNITE-13564:
-----------------------------------------
Summary: Improve SYSTEM_WORKER_BLOCKED reporting.
Key: IGNITE-13564
URL: https://issues.apache.org/jira/browse/IGNITE-13564
Project: Ignite
Issue Type: Improvement
Affects Versions: 2.8.1, 2.9
Reporter: Ivan Daschinskiy
Assignee: Ivan Daschinskiy
Fix For: 2.10
Currently, reporting of system thread blocking has major drawbacks.
1. As system worker blocking is detected by another thread, due to implementation, failure handler receives not full information about problem. In {{FailureContext}} we have only two fields -- {{type}} and {{err}}. Throwable {{err}} is generated in thread-detector flow, so we lost a context of main problem.
2. Currently, due to implementation, we print not full stacktrace of blocking thread in {{org.apache.ignite.internal.worker.WorkersRegistry#onIdle}}.
3. Current approach doesn't work when there is one thread in registry, this fact isn't checked and this can cause to infinite looping of single thread, calling {{onIdle}}
This two drawbacks can lead to completely loss of information about blocking system thread.
I suggests:
1. Add another parameter in {{FailureContext}}, namely {{worker}}
2. Fix threaddump printing.
3. Add assertion when there is only one system thread in registry
--
This message was sent by Atlassian Jira
(v8.3.4#803005)