[MCOL-1037] Race condition in FIFO buffer Created: 2017-11-16 Updated: 2020-08-25 Resolved: 2018-07-25 |
|
| Status: | Closed |
| Project: | MariaDB ColumnStore |
| Component/s: | None |
| Affects Version/s: | 1.0.10, 1.1.2, 1.1.3 |
| Fix Version/s: | 1.1.6 |
| Type: | Bug | Priority: | Major |
| Reporter: | Andrew Hutchings (Inactive) | Assignee: | Elena Kotsinova (Inactive) |
| Resolution: | Fixed | Votes: | 4 |
| Labels: | None | ||
| Attachments: |
mcol_1037.tar.gz
|
||||||||||||||||
| Issue Links: |
|
||||||||||||||||
| Epic Link: | ColumnStore Compatibility Improvements | ||||||||||||||||
| Sprint: | 2017-25, 2018-01, 2018-04, 2018-05, 2018-06, 2018-07, 2018-08, 2018-09, 2018-10, 2018-11, 2018-12, 2018-13, 2018-14, 2018-15 | ||||||||||||||||
| Description |
|
Race condition in FIFO buffer |
| Comments |
| Comment by Andrew Hutchings (Inactive) [ 2018-04-09 ] |
|
Richard found a user reporting that thread_resource_error::~thread_resource_error() was triggered. Could be that this exception is ignored on a thread processing a block of aggregates and therefore not counted. |
| Comment by Andrew Hutchings (Inactive) [ 2018-04-09 ] |
|
Tried to trigger an exception there, without the rest of the stack trace it is really hard to trigger at the right time (since we can get here numerous ways). ExeMgr aborted in every test I tried but it could be the wrong path. |
| Comment by Andrew Hutchings (Inactive) [ 2018-05-05 ] |
|
Assigned to Elena to look into. |
| Comment by Andrew Hutchings (Inactive) [ 2018-06-27 ] |
|
Both the ExeMgr issue and the potential root cause issue seem to point to TupleBPS::receiveMultiPrimitiveMessages. |
| Comment by Andrew Hutchings (Inactive) [ 2018-07-11 ] |
|
Race in FIFO buffer found using TSAN. After 25 runs with this patch I could not reproduce the bug. For QA: Please see my earlier notes about how to reproduce in AWS. Only a single UM/PM setup is needed without the extra config options. |
| Comment by Elena Kotsinova (Inactive) [ 2018-07-25 ] |
|
verified |