Bug 833202

Summary: transaction retries need to be cache aware
Product: Red Hat Enterprise Linux 6 Reporter: Nathan Kinder <nkinder>
Component: 389-ds-baseAssignee: Rich Megginson <rmeggins>
Status: CLOSED ERRATA QA Contact: Sankar Ramalingam <sramling>
Severity: unspecified Docs Contact:
Priority: medium    
Version: 6.4CC: jgalipea, jrusnack, nhosoi
Target Milestone: rc   
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
Fixed In Version: 389-ds-base- Doc Type: Bug Fix
Doc Text:
Cause: Having a directory server under a heavy update load. Consequence: Directory server crashes. Stack trace shows crash in libback-ldbm.so shared library in the ns-slapd process. Fix: Do not free internal entries that are in the cache when retrying failed transactions due to deadlock. Result: Directory server does not crash under a heavy update load.
Story Points: ---
Clone Of: Environment:
Last Closed: 2013-02-21 08:18:08 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:

Description Nathan Kinder 2012-06-18 21:07:12 UTC
This bug is created as a clone of upstream ticket:

The code that resets the original entries/mods/etc. when retrying deadlocked transactions in ldbm_back_delete et. al. need to be cache aware when deleting entries that are cached - for example, ldbm_delete.c:485
The problem is that e is cached, so cannot be freed like this.  Freeing cached entries seems to cause problems in the server, assertion failures, other crashes.

Comment 1 Jenny Severance 2012-06-19 18:36:13 UTC
Please add steps to verify this bug

Comment 2 Noriko Hosoi 2012-06-21 17:06:46 UTC
The attached test scripts and steps in this comment can be shared for verifying this bug.

After running the test, run search all.
ldapsearch -x ... -b "dc=example,dc=com" >> OUTPUT0

Restart the server and run search all, again.
ldapsearch -x ... -b "dc=example,dc=com" >> OUTPUT1

The 2 outputs OUTPUT0 and OUTPUT1 are supposed to be identical (to ensure the cache reflects the entries in DB).

Comment 3 Rich Megginson 2012-07-07 14:28:32 UTC
Also covered by the new test multi_plugin - the stress test
r6706 | rmeggins | 2012-07-07 08:24:23 -0600 (Sat, 07 Jul 2012) | 16 lines

Add multi_plugin test suite
This test enables and configures multiple plugins in the style
of ipa.  The stress version of this test will enable the 
TXN_TESTING in the server which will exercise the txn deadlock
and retry code paths in the server.
Tickets covered by this test:
Ticket #345 - db deadlock return should not log error
Ticket #387 - managed entry sometimes doesn't delete the managed entry
Ticket #360 - ldapmodify returns Operations error
Ticket #335 - transaction retries need to be cache aware
Bugs covered by this test:
830336 db deadlock return should not log error
830343 managed entry sometimes doesn't delete the managed entry
833218 ldapmodify returns Operations error
833202 transaction retries need to be cache aware

Comment 5 Ján Rusnačko 2013-01-29 11:30:54 UTC
On RHEL 6.4 Snapshot 5 with 389-ds-base-

multi_plugin startup 	100% (1/1)
multi_plugin run 	100% (6/6)
multi_plugin cleanup 	100% (1/1)


Comment 7 errata-xmlrpc 2013-02-21 08:18:08 UTC
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.