bereng commented on a change in pull request #870:
URL: https://github.com/apache/cassandra/pull/870#discussion_r568326365
##########
File path: src/java/org/apache/cassandra/dht/RangeStreamer.java
##########
@@ -353,9 +361,27 @@ public void addRanges(String keyspaceName,
ReplicaCollection<?> replicas)
*/
private boolean useStrictSourcesForRanges(AbstractReplicationStrategy
strat)
{
- return useStrictConsistency
- && tokens != null
- && metadata.getSizeOfAllEndpoints() !=
strat.getReplicationFactor().allReplicas;
+ boolean res = useStrictConsistency && tokens != null;
+
+ if (res)
+ {
+ int nodes = 0;
+
+ if (strat instanceof NetworkTopologyStrategy)
+ {
+ ImmutableMultimap<String, InetAddressAndPort> dc2Nodes =
metadata.getDC2AllEndpoints(snitch);
+
+ NetworkTopologyStrategy ntps = (NetworkTopologyStrategy) strat;
+ for (String dc : dc2Nodes.keySet())
+ nodes += ntps.getReplicationFactor(dc).allReplicas > 0 ?
dc2Nodes.get(dc).size() : 0;
+ }
+ else
+ nodes = metadata.getSizeOfAllEndpoints();
+
+ res = res && nodes > strat.getReplicationFactor().allReplicas;
Review comment:
I was thinking about branch prediction when writing that. Iirc from some
training an early return disables it so accumulating the result to return at
the end is preferable. You'd then want to keep `nodes` declaration inside the
branch to spare that cost when possible as well.
So I am removing the `res &&` bit indeed but the rest sgtm. Does it make
sense or am I being too far reached? This is not in the hot path iiuc so we may
want to favor readability :thinking:
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
