MySQL Bugs: #39235: NDB : No dropped signal handling for fragmented signals

Bug #39235	NDB : No dropped signal handling for fragmented signals
Submitted:	4 Sep 2008 9:59	Modified:	12 Dec 2008 13:42
Reporter:	Frazer Clement	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Cluster: Cluster (NDB) storage engine	Severity:	S3 (Non-critical)
Version:	5.1	OS:	Any
Assigned to:	Frazer Clement	CPU Architecture:	Any

Description:
When long signal buffer exhaustion in NDBD results in a signal being dropped, the usual handling mechanism does not take fragmented signals into account.
This could result in a crash as the fragmented signal handling would not know how to deal with the missing fragments.

How to repeat:
Send long fragmented signal that results in long signal buffer exhaustion.
Cases : 
  First fragment dropped, other fragments dropped : No problem
  First fragment dropped, other fragments arrive : Fragment assembly crash
  First fragment arrives, middle fragment(s) dropped, last fragment arrives : Fragment assembly succeeds with missing data
  First All fragments arrive except last : Fragment assembly record and long signal buffer leak.

Suggested fix:
Add fragmented signal handling to relevant SIGNAL_DROPPED_REP handling code.

Improved and tested patch with autotest testcase

Attachment: bug39235-with-test.patch (text/x-patch), 19.04 KiB.

i would probably wrap the Dbtc::testFragmentDrop in #ifdef ERROR_INSERT
maybe not so much for "optimization" but more for documentation of the code

also: i used error code 8074 (i think) so you need to shift the numbers...

comment on this: I actually sometimes reserve numbers in ERROR_CODES.txt
  (commit + push) *before* pushing actual patch, to avoid such clashes.

summary: ok to push, would be nice with the #ifdef, but not necessary

8074 appears to be unused so far.

I've added the #ifdefs round testFragmentDrop() definition + usage.

Will push to 6.4.

Pushed to 6.4.0

http://lists.mysql.com/commits/61110

Documented bugfix in the ndb-6.4.0 changelog as follows:

        When long signal buffer exhaustion in the ndbd process resulted
        in a signal being dropped, the usual handling mechanism ddid not
        take fragmented signals into account. This could result in a
        crash of the data node because the fragmented signal handling
        mechanism was not able to work with the missing fragments.