it seems this issue also happened on Impala3.3, not juse impala 3.2, but it's fixed in 3.3.
so, Cloudera support, how to fix this issue on imapla-3.2( CDH6.2.1), this issue is so critical cause many users encounter this issue and ask me what's happening, and i just can tell them this is a bug and i can't do anything.
may i ask another question ? if Impala3.2 can't resolve this issue ,may i update imapla 3,2 to impala 3.3 , i mean i don't update Cloudera Manager and CDH other components , just update impala .
i hope Cloudera Supporter can give me advises , thank you very much .
This generally happens when overwriting files in-place where Impala is still trying to read a cached version of the file. E.g. insert overwrite in Hive. So you can often avoid the problem if you can avoid doing that.
Otherwise doing a REFRESH of the table should resolve it.
i have found your workaround in JIRA, and will try. as you have mentioned REFRESH should resolve this issue, but actually it can't resolve, even i use "INVALIDATE METADATA" . that's why i open this session. anyway, thanks your solution.
My CDH env has been online more than half year, agent heartbeat timeout has never been happened, but after comparing the date of setting the impala parameter and agent heartbeat issue date , it seems there are connection, but I am not sure .
what I mean is the agent heartbeat timeout issue happened after I set the impala parameter "-max_cached_file_handles=0 ".