Skip to:

  1. Skip to Jira Navigation
  2. Skip to Side Navigation
  3. Skip to Main Content

Revert PS-3360 (Busy server prefers LRU flushing over flush list flushing too strongly)

Description

https://perconadev.atlassian.net/browse/PS-3360#icft=PS-3360 was developed when LRU flusher was performed by the page cleaners, before its flushing was split to separate threads. Now its logic is outdated and the comment invalid:

diff --git a/storage/innobase/buf/buf0flu.cc b/storage/innobase/buf/buf0flu.cc index c7672a6137e..443740ac200 100644 --- a/storage/innobase/buf/buf0flu.cc +++ b/storage/innobase/buf/buf0flu.cc @@ -3385,7 +3385,13 @@ DECLARE_THREAD(buf_flush_page_cleaner_coordinator)(                   n_flushed = n_processed_lru;   -               if (ut_time_ms() > next_loop_time) +               if (ut_time_ms() > next_loop_time +                   && ret_sleep != OS_SYNC_TIME_EXCEEDED) + +                       /* If our LRU flush took too long, skip the rest only +                       if the last iteration completed in time, otherwise we'd +                       be LRU flushing all the time, even if e.g. a sync flush +                       is waiting. */                         ret_sleep = OS_SYNC_TIME_EXCEEDED;                 else if (ret_sleep != OS_SYNC_TIME_EXCEEDED                     && srv_flush_sync

Analyze and revert.

Environment

None

has to be done before

Smart Checklist

Activity

Show:

Laurynas Biveinis February 12, 2018 at 4:18 AM

c9ab100f7fe50029bb2505c4780a99a600ed18c0

Done

Details

Assignee

Reporter

Labels

Time tracking

36m logged

Fix versions

Affects versions

Priority

Smart Checklist

Created February 12, 2018 at 4:15 AM
Updated March 2, 2018 at 2:27 PM
Resolved February 20, 2018 at 2:25 PM

Flag notifications