Proxy server ignores additional fragments on primary nodes
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Object Storage (swift) |
Fix Released
|
High
|
paul luse |
Bug Description
You can't get multiple fragments on the same node unless you rebalance the ring.
When you are doing rebalances - it's possible a node holding a fragment is not holding the correct fragment index according to it's current index in the ring.
If this happens the node that *is* holding the correct fragment index that belongs on that current primary will push it over to that node and it will end up holding multiple fragments.
However, the proxy will never consider multiple fragments from the same node, it's possible even that a fragment could be reconstructed to a primary node and the proxy could see multiples of the same indexes.
None of these extremely rare and not well understood failure conditions are handled gracefully in the proxy. We've been able to concoct some contrived failure modes that we know would result in 503s and 404s in the proxy.
Changed in swift: | |
assignee: | nobody → paul luse (paul-e-luse) |
Changed in swift: | |
importance: | Undecided → Critical |
importance: | Critical → High |
Changed in swift: | |
importance: | High → Critical |
Changed in swift: | |
importance: | Critical → High |
Fix proposed to branch: master /review. openstack. org/213147
Review: https:/