[
https://issues.apache.org/jira/browse/HBASE-18517?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vikas Vishwakarma updated HBASE-18517:
--------------------------------------
Description:
We had two cases now in our prod / pilot setups which is leading to humongous
log lines in RegionServer logs.
In one case one of the phoenix user had constructed a query with a really large
list of Id filters (61 MB) that translated into HBase scan that was running
slow which lead to responseTooSlow messages in the logs with the entire filter
list being printed in the logs, example
ipc.RpcServer - (responseTooSlow):
{"call":"Scan(org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ScanRequest)","starttimems":1501457864417,"responsesize":11,"method":"Scan","param":"region
{ type: REGION_NAME value: .....
org.apache.hadoop.hbase.filter.FirstKeyOnlyFilter\\022\351\\200\\036\\n(org.apache.phoenix.filter.SkipScanFilter
... <a really long filter list leading to single log lines of size 61MB> ...
There was another case where a use case had created a table with really large
key/region names. This was causing humongous log lines for flush and compaction
on these regions filling up the RS logs
These large logs usually cause issues with disk I/O load, loading the splunk
servers, even machine perf degradations. With 61 MB log lines basic log
processing commands like vim, scrolling the logs, wc -l , etc were getting
stuck. High GC activity was also noted on this cluster although not 100% sure
if it was related to above issue.
We should consider limiting the message size in logs which can be easily done
by adding a maximum width format modifier on the message conversion character
in log4j.properties
log4j.appender.console.layout.ConversionPattern=...: %m%n
to
log4j.appender.console.layout.ConversionPattern=...: %.10000m%n
was:
We had two cases now in our prod / pilot setups which is leading to humongous
log lines in RegionServer logs.
In one case one of the phoenix user had constructed a query with a really large
list of Id filters (61 MB) that translated into HBase scan that was running
slow which lead to responseTooSlow messages in the logs with the entire filter
list being printed in the logs, example
ipc.RpcServer - (responseTooSlow):
{"call":"Scan(org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ScanRequest)","starttimems":1501457864417,"responsesize":11,"method":"Scan","param":"region
{ type: REGION_NAME value: .....
org.apache.hadoop.hbase.filter.FirstKeyOnlyFilter\\022\351\\200\\036\\n(org.apache.phoenix.filter.SkipScanFilter
... <a really long filter list leading to single log lines of size 61MB> ...
There was another case where a use case had created a table with really large
key/region names. This was causing humongous log lines for flush and compaction
on these regions filling up the RS logs
These large logs usually cause issues with disk I/O load, loading the splunk
servers, even machine perf degradations. With 61 MB log lines basic log
processing commands like vim, scrolling the logs, wc -l , etc were getting
stuck. High GC activity was also noted on this cluster although not 100% sure
if it was related to above issue.
We should consider limiting the message size in logs which can be easily done
by adding a maximum width format modifier on the message conversion character
in log4j.properties
log4j.appender.console.layout.ConversionPattern=%d{ISO8601} %-5p [%t] %c{2}:
%m%n
to
log4j.appender.console.layout.ConversionPattern=%d{ISO8601} %-5p [%t] %c{2}:
%.10000m%n
> limit max log message width in log4j
> ------------------------------------
>
> Key: HBASE-18517
> URL: https://issues.apache.org/jira/browse/HBASE-18517
> Project: HBase
> Issue Type: Bug
> Reporter: Vikas Vishwakarma
> Assignee: Vikas Vishwakarma
>
> We had two cases now in our prod / pilot setups which is leading to humongous
> log lines in RegionServer logs.
> In one case one of the phoenix user had constructed a query with a really
> large list of Id filters (61 MB) that translated into HBase scan that was
> running slow which lead to responseTooSlow messages in the logs with the
> entire filter list being printed in the logs, example
> ipc.RpcServer - (responseTooSlow):
> {"call":"Scan(org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ScanRequest)","starttimems":1501457864417,"responsesize":11,"method":"Scan","param":"region
> { type: REGION_NAME value: .....
> org.apache.hadoop.hbase.filter.FirstKeyOnlyFilter\\022\351\\200\\036\\n(org.apache.phoenix.filter.SkipScanFilter
> ... <a really long filter list leading to single log lines of size 61MB> ...
> There was another case where a use case had created a table with really large
> key/region names. This was causing humongous log lines for flush and
> compaction on these regions filling up the RS logs
> These large logs usually cause issues with disk I/O load, loading the splunk
> servers, even machine perf degradations. With 61 MB log lines basic log
> processing commands like vim, scrolling the logs, wc -l , etc were getting
> stuck. High GC activity was also noted on this cluster although not 100% sure
> if it was related to above issue.
> We should consider limiting the message size in logs which can be easily done
> by adding a maximum width format modifier on the message conversion character
> in log4j.properties
> log4j.appender.console.layout.ConversionPattern=...: %m%n
> to
> log4j.appender.console.layout.ConversionPattern=...: %.10000m%n
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
