[LU-17250] Add new MDT to existing filesystem misses OST pools, nodemaps, and other config Created: 01/Nov/23 Updated: 30/Jan/24 |
|
| Status: | Open |
| Project: | Lustre |
| Component/s: | None |
| Affects Version/s: | Lustre 2.15.3 |
| Fix Version/s: | None |
| Type: | Bug | Priority: | Major |
| Reporter: | Andreas Dilger | Assignee: | Etienne Aujames |
| Resolution: | Unresolved | Votes: | 1 |
| Labels: | None | ||
| Issue Links: |
|
||||||||
| Severity: | 3 | ||||||||
| Rank (Obsolete): | 9223372036854775807 | ||||||||
| Description |
|
If new MDTs are added to an existing filesystem that has OST Pools configured, the new MDTs are missing all of the configuration parameters that describe the pools. In fact, the new MDTs will also be missing any other persistent config parameters that are stored in the per-target config logs by "lctl conf_param", "lctl pool_*", "lctl nodemap_*", etc. This issue does not apply to persistent parameters set with "lctl set_param -P", since this is a generic config log that is applied to all nodes and does not store specific configuration records into a per-target log. It probably makes sense for the MGS to process the config llog from another target (eg. fsname-MDT0000) to add these parameters to newly-added MDTs (and OSTs for that matter). The "regular" configuration records (attach, setup, add_uuid, new_profile, add_osc) should be skipped along with SKIP records, while conf_param, new_pool, add_pool, etc. records should be copied into the new target device log, with suitable replacements (eg. "s/MDT0000/MDTxxxx/"). |
| Comments |
| Comment by Andreas Dilger [ 01/Nov/23 ] |
|
I'm not sure whether it is better to exclude some logs and copy everything else, or to only include specific records and skip everything else? The former is more forward compatible and less likely to need constant attention, but the latter is a bit safer and would not unintentionally copy settings that are not needed. There would be some risk of copying a device-specific setting (eg. disable DoM on MDT0000), but that could be addressed afterward by cancelling that record from the llog. |
| Comment by Gerrit Updater [ 08/Jan/24 ] |
|
"Etienne AUJAMES <eaujames@ddn.com>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53614 |