[MCOL-1222] ColumnStore start/restart can return before system is ready Created: 2018-02-20 Updated: 2018-03-23 Resolved: 2018-03-23 |
|
| Status: | Closed |
| Project: | MariaDB ColumnStore |
| Component/s: | None |
| Affects Version/s: | 1.1.3 |
| Fix Version/s: | 1.1.4 |
| Type: | Bug | Priority: | Minor |
| Reporter: | Andrew Hutchings (Inactive) | Assignee: | Daniel Lee (Inactive) |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Sprint: | 2018-06 |
| Description |
|
As observed in the regression suite it is possible for mcsadmin (and maybe postConfigure) to return before ColumnStore is ready to accept queries. We probably shouldn't do this. |
| Comments |
| Comment by David Hall (Inactive) [ 2018-02-20 ] |
|
We should modify waitForActive() in mcsadmin to look at DBRM::getSystemQueryReady() |
| Comment by David Hill (Inactive) [ 2018-03-15 ] |
|
fixed... moved the waitForActive() from postConfigure and mcsadmin to oam APIs. added check for DBRM::getSystemQueryReady(). pull request https://github.com/mariadb-corporation/mariadb-columnstore-engine/pull/428 |
| Comment by David Hill (Inactive) [ 2018-03-15 ] |
|
pull request https://github.com/mariadb-corporation/mariadb-columnstore-engine/pull/428 |
| Comment by David Hill (Inactive) [ 2018-03-15 ] |
|
how to test... 1. postConfigure install and make sure system goes active and good return back to user |
| Comment by David Hall (Inactive) [ 2018-03-21 ] |
|
The new wait period of 10 seconds in a loop is excessive. The previous wait of 3 seconds in a loop is much less annoying to a user. |
| Comment by David Hill (Inactive) [ 2018-03-22 ] |
|
changing design to sleep for 3 seconds instead of 10 and print the dot to the screen every 9 seconds. this will keep the dots from rapping around on on startups, like multi-node systems. |
| Comment by Daniel Lee (Inactive) [ 2018-03-23 ] |
|
Build tested: 1.1.4-1 source /root/columnstore/mariadb-columnstore-server regression tested binary installation, restart system, shutdown and start system on a 1um2pm stack. |
| Comment by Daniel Lee (Inactive) [ 2018-03-23 ] |
|
Also tested concurrent queries right after system became available. |