Periodically flush info log out of application buffer #7488
Conversation
There was a problem hiding this comment.
@ajkr has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
|
@ajkr has updated the pull request. You must reimport the pull request before landing. |
There was a problem hiding this comment.
@ajkr has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Thanks for the fast review! Sorry I forgot to import so now need the internal diff stamped. The test failure looks likely fixed. |
This PR schedules a background thread (shared across all DB instances) to flush info log every ten seconds. This improves debuggability in case of RocksDB hanging since it ensures the log messages leading up to the hang will eventually become visible in the log. The bulk of this PR is moving monitoring/stats_dump_scheduler* to db/periodic_work_scheduler* and making the corresponding name changes since now the scheduler handles info log flushing, not just stats dumping.
|
@ajkr has updated the pull request. You must reimport the pull request before landing. |
There was a problem hiding this comment.
@ajkr has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Summary: This PR schedules a background thread (shared across all DB instances) to flush info log every ten seconds. This improves debuggability in case of RocksDB hanging since it ensures the log messages leading up to the hang will eventually become visible in the log. The bulk of this PR is moving monitoring/stats_dump_scheduler* to db/periodic_work_scheduler* and making the corresponding name changes since now the scheduler handles info log flushing, not just stats dumping. Pull Request resolved: facebook#7488 Reviewed By: riversand963 Differential Revision: D24065165 Pulled By: ajkr fbshipit-source-id: 339c47a0ff43b79fdbd055fbd9fefbb6f9d8d3b5
This PR schedules a background thread (shared across all DB instances)
to flush info log every ten seconds. This improves debuggability in case
of RocksDB hanging since it ensures the log messages leading up to the hang
will eventually become visible in the log.
The bulk of this PR is moving monitoring/stats_dump_scheduler* to db/periodic_work_scheduler*
and making the corresponding name changes since now the scheduler handles info
log flushing, not just stats dumping.