[SERVER-8835] _secondaryThrottle should GLE after every N documents rather than after each Created: 03/Mar/13 Updated: 09/Oct/18 Resolved: 01/Oct/18 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Sharding |
| Affects Version/s: | 2.4.0-rc1 |
| Fix Version/s: | None |
| Type: | Bug | Priority: | Major - P3 |
| Reporter: | Asya Kamsky | Assignee: | Alyson Cabral (Inactive) |
| Resolution: | Duplicate | Votes: | 0 |
| Labels: | None | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Issue Links: |
|
||||||||||||||||
| Operating System: | ALL | ||||||||||||||||
| Participants: | |||||||||||||||||
| Description |
|
The option _secondaryThrottle when set to true will GLE w:2 after every single document being copied/cloned. For cases of very small documents (aka very large number of documents in chunk range) when there are also inserts going on in the from shard into this range, this can be a lot slower (relative to unthrottled migration) than for a collection with fewer documents in a chunk. Maybe the waitForReplication() should be every N documents where N is some fraction of maxDocs in chunk (so say, waitForReplication() every 1/100th of chunk or 1/1000th). Otherwise the delay it introduces is too unpredictable. |
| Comments |
| Comment by Alyson Cabral (Inactive) [ 01/Oct/18 ] |
|
|
| Comment by Asya Kamsky [ 13/Jul/18 ] |
|
Are we removing secondary throttle as an option? Because the issue here is not slowing down or speeding up the balancer via range deleter tuning, this was specifically about secondary throttle on initial document copying being too wildly unpredictable depending on document size.
|
| Comment by Gregory McKeon (Inactive) [ 12/Jul/18 ] |
|
We won't fix _secondaryThrottle, but there's a configuration for slowing down the balancer described in |
| Comment by Asya Kamsky [ 19/Apr/18 ] |
|
|
| Comment by Asya Kamsky [ 19/Apr/18 ] |
|
This ticket is to improve secondary throttle parameter's impact on performance, even though we no longer turn it on by default, it's still not useful for collections with small documents (aka chunks with a very large number of documents). |
| Comment by Kaloian Manassiev [ 20/Jan/16 ] |
|
Resolving in place of |
| Comment by Eliot Horowitz (Inactive) [ 01/Apr/13 ] |
|
without more clear evidence one way or the other, don't want to change |