Ticket #305 (reopened defect)

Opened 14 months ago

Last modified 5 days ago

LOG service record write throughput is very bad

Reported by: hafe Owned by: hafe
Priority: major Milestone: PL 3.0.2
Component: LOGSv Version: 2.0.0
Keywords: Cc:
patch waiting for maintainer: no

Description

The reason is that the O_SYNC flag is set when opening a file. When the logsv directory is using drbd is gets even worse.

The suggestion is to removed the O_SYNC flag. The consequence is that a controller power off could result in loss of acknowledged log records in the log file.

Removing the O_SYNC flag results in ~250 times improvement. From 20 to 5000 records per second!

Attachments

Change History

Changed 13 months ago by hafe

  • status changed from new to closed
  • resolution set to fixed

The O_SYNC flag was removed.

Changed 5 days ago by hafe

  • status changed from closed to reopened
  • resolution fixed deleted
  • milestone changed from PL 2.0.2 to PL 3.0.2

The O_SYNC flag was unfortunately reintroduced in 3.0 during a merge.

Another problem when calling close(), EINTR is not handled properly. Have seen this!

Add/Change #305 (LOG service record write throughput is very bad)

Author



Action
as reopened
Note: See TracTickets for help on using tickets.