cassandra metric CrossNodeLatency的p98时延较大,导致平均时延基本在10ms以上，经常出现800ms以上的，max有1.6s的情况。排查节点间的网络时延良好，可能还与什么因素有关？GC？还有什么因素
I don't have a complete list but 2 things which come to mind are:
- check for blips in network which could cause spikes in latency
- review queries and look for large partitions
The cross-node latency metric is a measure of how long it took for a message response to be returned to a note. The message itself contains a timestamp of when it was sent so the receiving node is able to calculate the latency. Network jitter although intermittent can certainly contribute to spikes in latency.
Additionally, large payloads (mutations larger than usual; large read result set) will also contribute to increased latency. Cheers!
6 People are following this question.
