Swift/Deploy Plan - R720xds in tampa
From Wikitech
< Swift(Difference between revisions)
ArielGlenn (Talk | contribs) (→Proposed schedule) |
ArielGlenn (Talk | contribs) |
||
| Line 12: | Line 12: | ||
Pushing our luck a little bit, we can go from 33 weight for the object rings to 66 in 3 days and from 66 to 100 in another 4. This is a compressed schedule, it really doesn't let network traffic to the new hosts settle down completely, but it will let us get the new boxes in and the old ones shipped. | Pushing our luck a little bit, we can go from 33 weight for the object rings to 66 in 3 days and from 66 to 100 in another 4. This is a compressed schedule, it really doesn't let network traffic to the new hosts settle down completely, but it will let us get the new boxes in and the old ones shipped. | ||
| − | We should do these | + | We should do these one at a time, it's taking too long for new boxes to get caught up. Schedule for first deployment in cluster with 33% weight: |
| − | *ms-be3 | + | *ms-be3 on Mon Dec 3 |
| − | *ms-be5 | + | *ms-be5 on Dec 5 (maybe, we'll see if two days is enough) |
| + | *ms-be11 on Dec 7 | ||
| + | *ms-be12 on Dec 10 | ||
Replace these but turn them into ceph hosts: | Replace these but turn them into ceph hosts: | ||
| − | *me-be9 | + | *me-be9 on Dec 12 |
| − | *me be-2 | + | *ms-be1 on Dec 14 |
| + | *me be-2 on Dec 17 | ||
| + | *ms ms-be4 on Dec 19 | ||
This schedule will be adjusted (shorter or longer) as we see how the cluster behaves. | This schedule will be adjusted (shorter or longer) as we see how the cluster behaves. | ||
Revision as of 16:38, 3 December 2012
Draft, edit at will
Current status
- Currently replaced: ms-be6,7,8,10
- Host which has fallen over a few times recently: ms-be3
- Hosts with disk issues (so they should be replaced next): ms-be5,11,12
- Other hosts with ssds: ms-be9
- The rest: ms-be1,2,4
Proposed schedule
Pushing our luck a little bit, we can go from 33 weight for the object rings to 66 in 3 days and from 66 to 100 in another 4. This is a compressed schedule, it really doesn't let network traffic to the new hosts settle down completely, but it will let us get the new boxes in and the old ones shipped.
We should do these one at a time, it's taking too long for new boxes to get caught up. Schedule for first deployment in cluster with 33% weight:
- ms-be3 on Mon Dec 3
- ms-be5 on Dec 5 (maybe, we'll see if two days is enough)
- ms-be11 on Dec 7
- ms-be12 on Dec 10
Replace these but turn them into ceph hosts:
- me-be9 on Dec 12
- ms-be1 on Dec 14
- me be-2 on Dec 17
- ms ms-be4 on Dec 19
This schedule will be adjusted (shorter or longer) as we see how the cluster behaves.
Note: all hosts get ssds
Logistics
- What is our deadline for returning the C2100s?
- Do we have enough ssds to put in all remaining 720xds? If not let's get them (and what's the turnaround time for those?) YES we do.
