Todd Lipcon created KUDU-1616:
---------------------------------
Summary: Client should cache non-existence of a table
Key: KUDU-1616
URL: https://issues.apache.org/jira/browse/KUDU-1616
Project: Kudu
Issue Type: Bug
Components: client
Affects Versions: 1.0.0
Reporter: Todd Lipcon
I was running YCSB and dropped the table without stopping the client. The load
on the master quickly shot up because the client was getting "not found" errors
and trying to query the master for locations. Every op was immediately failing,
but because YCSB doesn't exit on failure, it kept generating new ops at the
same rate that the master could reject the location requests. This resulted in
50k+/sec RPC/sec against the master.
The master didn't fall over, but if there were a lot of clients sending 50k/sec
in a larger cluster, we'd probably start impacting requests against good tables.
To solve this, the client should have a negative cache of deleted table ids.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
