Ticket #108 (closed defect: fixed)

Opened 1 year ago

Last modified 2 months ago

midcom's rcs scripts freeze system

Reported by: piotras Assigned to: rambo
Priority: blocker Milestone: 1.8 Thor
Component: MidCOM core Version:
Keywords: Cc:

Description (Last modified by bergie)

It's more rcs issue than midcom itself. I can find plenty of such messages in apache error log repeated "endlessly": 

co: writable /var/lib/midgard/rcs/8c5ea2aa0e7711dc8cdc83a666378db78db7 exists; checkout aborted/var/lib/midgard/rcs/8c5ea2aa0e7711dc8cdc83a666378db78db7,v  <--  /var/lib/midgard/rcs/8c5ea2aa0e7711dc8cdc83a666378db78db7
/var/lib/midgard/rcs/8c5ea2aa0e7711dc8cdc83a666378db78db7,v  -->  /var/lib/midgard/rcs/8c5ea2aa0e7711dc8cdc83a666378db78db7 
revision 1.70240 (locked)

I think , nice description of the problem can be found here: http://www.oreilly.com/catalog/rcs/chapter/ch03.html

Disabling midcom's rcs seems to solve issue of "frozen" system.

Change History

08/23/07 15:25:08 changed by bergie

  • owner changed from bergie to tarjei.
  • component changed from Asgard to MidCOM core.
  • description changed.

08/31/07 14:30:02 changed by tarjei

  • status changed from new to assigned.

Hmm, I'm not sure where this happens, but we obviously need to add unlocking to midcom. I'm not sure where yet though.

% rcs -u xform.c RCS file: xform.c,v 1.1 unlocked done

10/05/07 10:31:05 changed by bergie

  • priority changed from major to blocker.

Raising priority as this seems to be popping up more frequently.

10/05/07 14:39:16 changed by bergie

(In [12676]) Log the RCS commands used to help debugging, refs #108

12/17/07 10:26:14 changed by bergie

  • owner changed from tarjei to rambo.
  • status changed from assigned to new.

12/17/07 10:28:21 changed by bergie

Some actions to do here:

  • Fix the RCS commands to perform locking/unlocking properly
  • Split the RCS directory to subdirectories from first two letters of a GUID (like the BlobDir? works)

12/17/07 13:41:07 changed by rambo

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

Revisions r13921 r13923 and r13924 are related to this. AFAIUnderstood from the linked document the unlocking is not neccessary unless we want to discard changes (ci unlocks), the messages just went to stderr (perhaps incorrectly) and were too verbose plus slightly confusing.

Most serious issue with RCS in general is that likely some objects that should not use RCS do not actually disable it (and DBA does it by default unless disabled for object), or that some cron or such touches a ton of objects causing huge number of shells to be opened for the RCS operations (which is expensive...).

Also a related issue fixed earlier was attachment serving opening attachments in write mode, causing unneccessary touch (and RCS updates) on each attachment serving.

If more RCS related issues rise, open more specific tickets (like "Object xx uses RCS even though it probably should not", or "RSS feed update always touches all articles even if they're not modified").

12/17/07 13:47:14 changed by rambo

Oh yeah, forgot to mention: to keep your revision histories after this use http://yousite.example.com/midcom-exec-midcom/move_legacy_rcs.php -script to move the old RCS files to the new blobdir-like hierarchy.

09/30/08 15:11:19 changed by bergie

  • milestone changed from MidCOM 2.8 to 1.8 Thor.