CASSANDRA-20323 Avoid limit on RFP fetch in the case of an unresolved static row #3916
Conversation
patch by Caleb Rackliffe; reviewed by ? for CASSANDRA-20323
| // provide a limit. (In the unresolved static case, we have no way of knowing how many stale rows we might | ||
| // read on a silent replica before finding a live one.) | ||
| ClusteringIndexFilter filter = unresolvedStatic ? command.clusteringIndexFilter(key) | ||
| : new ClusteringIndexNamesFilter(clusterings, command.isReversed()); |
There was a problem hiding this comment.
This is the crux of the patch. I went back and forth a bit, but I don't know if we can avoid fetching the rest of the partition (in the current data range). In practice, we're not going to be using ridiculously small fetch sizes, so I don't know how much of a problem this will actually be, but the viral nature of static columns is what it is.
There was a problem hiding this comment.
The only way I can imagine improving this is perhaps paging so that we don't fetch every row from a very large partition at once. Not 100% sure how to do that yet, but looking around the code to try to find some example of doing this w/ an UnfilteredPartitionIterator
maedhroz
changed the title
| String select = withKeyspace("SELECT pk0, ck0, ck1, s0, s1 FROM %s.single_predicate WHERE ck1 = 2 ALLOW FILTERING"); | ||
| assertRows(cluster.coordinator(1).execute(select, ConsistencyLevel.ALL), row(0, 1, 2, 6, 7)); | ||
| } | ||
|
|
There was a problem hiding this comment.
Moved to ReplicaFilteringWithStaticsTest
| String select = withKeyspace("SELECT pk0, ck0, ck1, s0, s1 FROM %s.single_predicate WHERE ck1 = 2 ALLOW FILTERING"); | ||
| assertRows(CLUSTER.coordinator(1).execute(select, ConsistencyLevel.ALL), row(0, 1, 2, 6, 7)); | ||
| } | ||
|
|
There was a problem hiding this comment.
Moved to ReplicaFilteringWithStaticsTest
patch by Caleb Rackliffe; reviewed by ? for CASSANDRA-20323