projects / ceph.git / blame
| Commit | Line | Data |
|---|---|---|
| 9f95a23c TL |
1 | OSD Support Module |
| 2 | ================== | |
| 3 | The OSD Support module holds osd specific functionality that | |
| 4 | is needed by different components like the orchestrators. | |
| 5 | ||
| 6 | In current scope: | |
| 7 | ||
| 8 | * osd draining | |
| 9 | ||
| 10 | Enabling | |
| 11 | -------- | |
| 12 | When an orchestrator is used this should be enabled as a dependency. | |
| 13 | (*currently only valid for the cephadm orchestrator*) | |
| 14 | ||
| 15 | The *osd_support* module is manually enabled with:: | |
| 16 | ||
| 17 | ceph mgr module enable osd_support | |
| 18 | ||
| 19 | Commands | |
| 20 | -------- | |
| 21 | ||
| 22 | Draining | |
| 23 | ######## | |
| 24 | ||
| 25 | This mode is for draining OSDs gracefully. `Draining` in this context means gracefully emptying out OSDs by setting their | |
| 26 | weight to zero. An OSD is considered to be drained when no PGs are left. | |
| 27 | ||
| 28 | :: | |
| 29 | ||
| 30 | ceph osd drain $osd_id | |
| 31 | ||
| 32 | Takes a $osd_id and schedules it for draining. Since that process can take | |
| 33 | quite some time, the operation will be executed in the background. To query the status | |
| 34 | of the operation you can use: | |
| 35 | ||
| 36 | :: | |
| 37 | ||
| 38 | ceph osd drain status | |
| 39 | ||
| 40 | This gives you the status of all running operations in this format:: | |
| 41 | ||
| 42 | [{'osd_id': 0, 'pgs': 1234}, ..] | |
| 43 | ||
| 44 | If you wish to stop an OSD from being drained:: | |
| 45 | ||
| 46 | ceph osd drain stop [$osd_id] | |
| 47 | ||
| 48 | Stops all **scheduled** osd drain operations (not the operations that have been started already) | |
| 49 | if no $osd_ids are given. If $osd_ids are present it only operates on them. | |
| 50 | To stop and reset the weight of already started operations we need to save the initial weight | |
| 51 | (see 'Ideas for improvement') | |
| 52 | ||
| 53 | ||
| 54 | Ideas for improvement | |
| 55 | ---------------------- | |
| 56 | - add health checks set_health_checks | |
| 57 | - use objects to represent OSDs | |
| 58 | - allows timestamps, trending information etc | |
| 59 | - save osd drain state (at least the osd_ids in the mon store) | |
| 60 | - resume after a mgr crash | |
| 61 | - save the initial weight of a osd i.e. (set to initial weight on abort) |