Description
When opening several buckets in a row, one of which is tried with
wrong password in the middle, the opening of last buckets can timeout.
During investigation it was seen that in a goodA/wrong/goodB bucket open
sequence, the CarrierRefresher would schedule a GetBucketConfigRequest
for "goodA" and it would go into a long-lasting retry loop.
Upon investigating further, it appears this is due to the ongoing
attempt at opening "wrong", which puts the Node(s) into DEGRADED mode.
Since the KeyValueLocator won't dispatch a message to a node that isn't
CONNECTED, this goes directly into the retry loop even though the
request could go through (the corresponding bucket's services are
CONNECTED).
Attachments
Issue Links
- is duplicated by
-
JVMCBC-353 Open good/bad/good bucket cascade leads to weird state
- Resolved
- relates to
-
JVMCBC-353 Open good/bad/good bucket cascade leads to weird state
- Resolved