[SERVER-32982] Remove rolled back index builds after rollback to a checkpoint Created: 29/Jan/18  Updated: 16/Feb/18  Resolved: 16/Feb/18

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

Type: Bug Priority: Major - P3
Reporter: Judah Schvimer Assignee: William Schultz (Inactive)
Resolution: Duplicate Votes: 0
Labels: rollback-functional
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Depends
depends on SERVER-29933 Roll Back to Checkpoint: Notify subsy... Closed
Duplicate
duplicates SERVER-33359 Have RTT storage engines manage rolli... Closed
Operating System: ALL
Sprint: Repl 2018-02-26
Participants:

 Description   

On primaries, we timestamp the beginning of an index build with a timestamp before the "createIndexes" oplog entry. Thus after we roll back to a checkpoint, the index with ready: false may still be around. If we crash and restart, WT will then rebuild the index, even though it doesn't exist on some nodes.



 Comments   
Comment by Daniel Gottlieb (Inactive) [ 16/Feb/18 ]

Oops, I forgot this ticket existed. We're changing the solution to push this into storage documented in SERVER-33359. That solution requires a little help from replication, but most importantly no longer relies on op observers and changing when replication recovery truncates the oplog.

Comment by Judah Schvimer [ 12/Feb/18 ]

This can likely be done in conjunction with SERVER-29933.

Comment by Judah Schvimer [ 30/Jan/18 ]

This can result in inconsistent index definitions across nodes.

Comment by Spencer Brody (Inactive) [ 30/Jan/18 ]

judah.schvimer Does this result in inconsistent index definitions across nodes? Or just in unnecessary work done to build the index which is then immediately thrown out?

Generated at Thu Feb 08 04:31:55 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.