I was looking at ResetTSProxy() in the context of AddServer async task yesterday, it is possible that the missing leader transient error case handling could be racy. Before providing the accurate problem description, I will try to repro the issue with a test next week and file/fix a JIRA if there is a genuine race.
Tx, On Wed, Mar 8, 2017 at 9:42 PM, Todd Lipcon <[email protected]> wrote: > Moving thread to dev@ for a follow-up question: can we make this error > message clearer, if this is an expected behavior? I'm a little confused as > to what sequence of events leads to this, though. > > On Mon, Mar 6, 2017 at 6:57 PM, <[email protected]> wrote: > > > Hi Paul > > > > That somewhat misleading message means that that particular tablet > > doesn't have (and never had) a leader. > > It might be a transient error, the master will continue to try the > alter > > but you might want to check if you have crashed servers and/or > > under-replicated tablets. > > > > David > > > > Sent from my iPhone > > > > > On Mar 6, 2017, at 6:16 PM, Paul Brannan <[email protected]> > > wrote: > > > > > > Using Kudu 1.2, I'm trying to add a range partition to a table, and it > > appears to succeeded (Alter() returns Status::OK()), but I see this in > > kudu-master.WARNING: > > > > > > W0306 21:03:42.175981 54035 catalog_manager.cc:2394] Unable to reset TS > > proxy: Not found: No leader found: Tablet b9aecb10e5e844b885a753c615432d > f9 > > has no consensus state > > > W0306 21:03:42.175994 54035 catalog_manager.cc:3005] Failed to send > > alter table request: Not found: Failed to reset TS proxy: No leader > found: > > Tablet b9aecb10e5e844b885a753c615432df9 has no consensus state > > > > > > What does this mean and what might I have done wrong? > > > > > > > > > -- > Todd Lipcon > Software Engineer, Cloudera >
