[SERVER-16640] Long running queries on the secondary with WT Created: 22/Dec/14  Updated: 15/Jan/15  Resolved: 29/Dec/14

Status: Closed
Project: Core Server
Component/s: Replication
Affects Version/s: None
Fix Version/s: 2.8.0-rc5

Type: Task Priority: Major - P3
Reporter: Kaloian Manassiev Assignee: Kaloian Manassiev
Resolution: Done Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Related
related to SERVER-16667 Isolated writes with extreme latencie... Closed
Backwards Compatibility: Fully Compatible
Participants:

 Description   

We have disabled yielding with WT, because it is not necessary due to the use of snapshots.

On secondaries though, because of the way the Parallel Batch Writer lock is implemented, it is possible that a long running query would prevent the log applier threads from starting until it is completed and thus cause backlog.

This ticket is to verify this behaviour and if it turns out to be true to figure out a solution.



 Comments   
Comment by Kaloian Manassiev [ 29/Dec/14 ]

Author:

{u'username': u'kaloianm', u'name': u'Kaloian Manassiev', u'email': u'kaloian.manassiev@mongodb.com'}

Message: SERVER-16431 Enable yielding for all storage engines
Branch: master
https://github.com/mongodb/mongo/commit/8f83ff2335ea0e2dd0dc2fc79b6509c4f8f2d4fc

Comment by Andy Schwerin [ 24/Dec/14 ]

Lack of yielding helped expose SERVER-16651, or more pessimistically, exacerbated it. Essentially, without yielding, long-running operations cause all acquisitions of the global lock in MODE_S or MODE_X to potentially block future read and write operations to individual collections. I recommend bringing yielding behavior into doc-locking storage engines. The underlying mechanisms are already present, anyways, since locks must be released between getMore calls in the current implementation of mongod.

Generated at Thu Feb 08 03:41:46 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.