A kill -9 kills the process without giving the process any notifications
nor any chance to perform cleanup or continue with an orderly shutdown.
I usually just issue a kill to the PID and let ES come down on its own. I
haven't had any problems, but I don't hurry it along either.
I currently have 4 indices: One tiny (for testing mappings), one relatively
small (26K documents), and two very large (96M documents in 5 shards, and
76M documents in 16 shards). No replicas at this time.
During start-up, I have a waitForYellow with a 5-minute timeout (after
which it refuses to start subsequent applications), and have never seen it
fail nor take more than perhaps 10 to 20 seconds.
After a full cycle of erase+create followed by a full bulk-load, there is
some amount of indexing and flushing that occurs after the bulk-load
process itself completes. But I'm not interested in shutting it down right
after bulk-loading, so by the time I do a shutdown via kill it only seems
to take 3 to 6 seconds before I can start it up again.
If I try to start too soon, it complains about a lock on the data
directory. So I would just wait a bit more and then it started OK. I'm more
patient now.
But I've never seen any data loss nor corruption. And there are some
instances that were originally pre-loaded using ES 19.4, and were
subsequently queried and updated using ES 19.4, 19.10, and now 20.4. No
issues at all. Very smooth migration!
A kill -9 kills the process without giving the process any notifications
nor any chance to perform cleanup or continue with an orderly shutdown.
I usually just issue a kill to the PID and let ES come down on its own. I
haven't had any problems, but I don't hurry it along either.
I currently have 4 indices: One tiny (for testing mappings), one
relatively small (26K documents), and two very large (96M documents in 5
shards, and 76M documents in 16 shards). No replicas at this time.
During start-up, I have a waitForYellow with a 5-minute timeout (after
which it refuses to start subsequent applications), and have never seen it
fail nor take more than perhaps 10 to 20 seconds.
After a full cycle of erase+create followed by a full bulk-load, there is
some amount of indexing and flushing that occurs after the bulk-load
process itself completes. But I'm not interested in shutting it down right
after bulk-loading, so by the time I do a shutdown via kill it only seems
to take 3 to 6 seconds before I can start it up again.
If I try to start too soon, it complains about a lock on the data
directory. So I would just wait a bit more and then it started OK. I'm more
patient now.
But I've never seen any data loss nor corruption. And there are some
instances that were originally pre-loaded using ES 19.4, and were
subsequently queried and updated using ES 19.4, 19.10, and now 20.4. No
issues at all. Very smooth migration!
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.