Please go through this section: http://kafka.apache.org/documentation.html#basic_ops_restarting
On Tue, Aug 19, 2014 at 01:19:34AM +0400, Ivan Balashov wrote: > Hi, > > Sorry if this had been answered before. Although, I couldn't find any > information besides "controlled shutdown of broker", which, I believe > not fully applies here. > > Could anyone suggest what would be the safest strategy to shut down > kafka cluster? Should brokers be brought down one-by-one or > simultaneously? Does this strategy change depend on whether > "controlled shutdown" is on? > > According to my brief observation, ad-hoc broker shutdown (mix and > match INT, TERM and KILL signals to the most resilient brokers) > results in prolonged cluster start up, with a long chain of > "Recovering unflushed segment 0" seen on some brokers. Is it expected > to see these after each shutdown/restart? > > Thanks in advance,
