Implement repair queue #692
Conversation
morimoto-cybozu
force-pushed
the
repair-queue
branch
13 times, most recently
from
7f482c6 to
e1841a9
Compare
morimoto-cybozu
changed the title
morimoto-cybozu
force-pushed
the
repair-queue
branch
from
e1841a9 to
3105433
Compare
morimoto-cybozu
force-pushed
the
repair-queue
branch
from
3105433 to
ffa81f4
Compare
morimoto-cybozu
force-pushed
the
repair-queue
branch
from
ffa81f4 to
ce4fd3f
Compare
| "address": entry.Address, | ||
| log.FnError: err, | ||
| }) | ||
| entry.Status = cke.RepairStatusProcessing |
There was a problem hiding this comment.
In the reboot queue, the "queued" status is set when drain is failed.
Why is processing status used in repaire queue?
https://github.com/cybozu-go/cke/blob/v1.27.3/op/reboot.go#L552
There was a problem hiding this comment.
@masa213f
I introduced the processing status for the repair queue to separate (a) the status of a queue entry and (b) the status of a step in an entry.
It would be confusing to manage both (a) and (b) in one field.
Each repair queue entry is designed to have multiple repair steps.
Currently we have three types of step statuses, waiting, draining, and watching.
When processing an entry, we should manage the loop of waiting, draining, and watching.
Furthermore, we may add new step statuses to extend the repair step's ability.
It would be hard to manage the transition diagram of the mixture of the entry statuses and the step statuses while increasing the types of step statuses.
I prefer to separate the entry statuses and the step statuses for simplicity.
There was a problem hiding this comment.
Thank you for the description. I understood.
When a node's repair waits for the drain in step 2, the CKE will execute the other node's repair.
And then, more entries than "max_concurrent_repairs" will be in the "processing" status.
Signed-off-by: morimoto-cybozu <[email protected]>
Signed-off-by: morimoto-cybozu <[email protected]>
Signed-off-by: morimoto-cybozu <[email protected]>
Signed-off-by: morimoto-cybozu <[email protected]>
Signed-off-by: morimoto-cybozu <[email protected]>
morimoto-cybozu
force-pushed
the
repair-queue
branch
from
cfa2b32 to
60afa8c
Compare
Signed-off-by: morimoto-cybozu [email protected]