[LU-9789] Create JobID prefix Created: 20/Jul/17 Updated: 21/Jan/20 Resolved: 02/Dec/19 |
|
| Status: | Closed |
| Project: | Lustre |
| Component/s: | None |
| Affects Version/s: | None |
| Fix Version/s: | None |
| Type: | Improvement | Priority: | Minor |
| Reporter: | Ben Evans (Inactive) | Assignee: | Ben Evans (Inactive) |
| Resolution: | Duplicate | Votes: | 0 |
| Labels: | None | ||
| Issue Links: |
|
||||||||
| Severity: | 3 | ||||||||
| Rank (Obsolete): | 9223372036854775807 | ||||||||
| Description |
|
Create JobID prefix to distinguish among clusters, or specific nodes within clusters (such as login nodes). If a site has a single shared filesystem and multiple clusters/schedulers, it could be possible for a JobID to be insufficiently unique, or difficult to determine where the job ran. Adding a small prefix would allow jobstats consumers a better understanding of where a job was run. |
| Comments |
| Comment by Gerrit Updater [ 21/Sep/17 ] |
|
Ben Evans (bevans@cray.com) uploaded a new patch: https://review.whamcloud.com/29150 |
| Comment by Andreas Dilger [ 27/Sep/17 ] |
|
Ben, you may also want to work with qian to improve the NRS TBF wildcard support, so that it is possible to specify NRS rules based on the Cluster ID. Another option is to add the cluster ID to the nodemap, so that this is prefixed to the JobID for all clients, regardless of whether they set this correctly on the client or not. That also avoids the issue of stuffing a long cluster ID prefix into the limited JobID string in the RPC. It would only need to be added when the JobID is printed on the server. Qian, Sebastien, it may also be useful to improve NRS so that rules can be set on a per-nodemap basis, but that should be done in the context of a different LU ticket. |
| Comment by Gerrit Updater [ 13/Apr/18 ] |
|
Ben Evans (bevans@cray.com) uploaded a new patch: https://review.whamcloud.com/31995 |
| Comment by Ben Evans (Inactive) [ 02/Dec/19 ] |
|
Duplicate of |