Bug 240897

(In reply to comment #10)
> BZ Comment #4 Step #8
> 
> Anhy idea why the indexing hung at this stage? I believe this was the root
> cause of all prblems that happened with Hyperion Solutions. It would comes
> as below.

It would seem that there is a bug with creating a browsing index on-line.

> 
> - One admin ran browsing index.
> - It hung. Don't know the reason for the hang, but it's clear from Step
> #8 of Comment #4 that indexing may hang due to "some" reasons.
> - Another admin found database in read-only mode. (At Hyperion more than
> one person has admin access, I guess)
> - Unchecked it. Didn't help. Killed it. Indexing got corruped and found
> an OU missing.
> 
> I see the possibility for another bugzilla which could come from Step #8
> of comment #4.
> 
> Thought?

Yes.  The server should listen for close events and respond.  It seems that in
this case indexing does not respond to the close event.

> 
> --Sadique
> 
> Version set to: '7.1'
> 
> This event sent from IssueTracker by sputhenp 
>  issue 121430
>               

Created attachment 200321 [details]
cvs diffs

Problem analysis:
[..] - Waiting for 4 database threads to stop
[..] - All database threads now stopped
[..] - libdb: DB_TXN->abort: Log undo failed for LSN: 3
1282125: No such file or directory
[..] - libdb: PANIC: No such file or directory
[..] - libdb: Unable to abort transaction 0x8000004b: DB_RUNRECOVERY: Fatal
error, run database recovery
[..] - libdb: PANIC: DB_RUNRECOVERY: Fatal error, run database recovery
[..] - libdb: Error: closing the transaction region with active transactions
[..] - slapd stopped.
Browsing index creation is implemented using task, which spawns a dedicated
task thread to do the task (index creation in this case).  When shutting down,
the server counts the active thread count and when it reaches 0, it closes the
backend db.  The problem here was the task threads were not counted in the
active thread count.  Therefore, the server closed the backend db while the
reindex task thread was still accessing the db index file.

Fix description:
1. moved active thread count from ns-slapd to libslapd for the task threads to
use.
2. provided APIs to increment/decrement/get the active thread count
3. let task threads increment the active thread count when it's spawned and
decrement it when it quits.

Files: ldapserver/ldap/server/slapd/
	    libglobs.c, configdse.c, conntable.c
	    proto-slap.h, connection.c, daemon.c
	    task.c, monitor.c, psearch.c
	    fe.h, globals.c

I think active_threads needs to be PRInt32 and it should be static.  I think you
can cast it to int in g_decr_active_threadcnt()

Created attachment 200931 [details]
cvs diff libglobs.c

(In reply to comment #16)
> I think active_threads needs to be PRInt32 and it should be static.  I think
you
> can cast it to int in g_decr_active_threadcnt()
Thank you for your comments, Rich.  I've updated libglobs.c.

And this is the correct file list:
Files: ldapserver/ldap/server/slapd/
	    libglobs.c,
	    proto-slap.h, connection.c, daemon.c
	    task.c, monitor.c, psearch.c
	    fe.h, globals.c

Created attachment 200941 [details]
cvs commit message

Reviewed by Rich (Thank you!)

Checked in into HEAD.