You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Pranav Murugappan (JIRA)" <ji...@apache.org> on 2017/10/24 10:10:00 UTC

[jira] [Updated] (SOLR-11233) GC_LOG_OPTS customisation is a little confusing

     [ https://issues.apache.org/jira/browse/SOLR-11233?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Pranav Murugappan updated SOLR-11233:
-------------------------------------
    Attachment: SOLR-11233.patch

# SOLR-11233
## GC_LOG_OPTS customisation is a little confusing

https://issues.apache.org/jira/browse/SOLR-11233


Quoting from the recent release - https://github.com/apache/lucene-solr/blob/releases/lucene-solr/7.1.0/solr/bin/solr#L1801

```bash

1781 # Establish default GC logging opts if no env var set (otherwise init to sensible default)
1782 if [ -z ${GC_LOG_OPTS+x} ]; then
1783   if [[ "$JAVA_VER_NUM" < "9" ]] ; then
1784     GC_LOG_OPTS=('-verbose:gc' '-XX:+PrintHeapAtGC' '-XX:+PrintGCDetails' \
1785                  '-XX:+PrintGCDateStamps' '-XX:+PrintGCTimeStamps' '-XX:+PrintTenuringDistribution' \
1786                  '-XX:+PrintGCApplicationStoppedTime')
1787   else
1788     GC_LOG_OPTS=('-Xlog:gc*')
1789   fi
1790 else
1791   GC_LOG_OPTS=($GC_LOG_OPTS)
1792 fi
1793
```

```bash 

1794 # if verbose gc logging enabled, setup the location of the log file and rotation
1795 if [ "$GC_LOG_OPTS" != "" ]; then
1796   if [[ "$JAVA_VER_NUM" < "9" ]] ; then
1797     gc_log_flag="-Xloggc"
1798     if [ "$JAVA_VENDOR" == "IBM J9" ]; then
1799       gc_log_flag="-Xverbosegclog"
1800     fi
1801     GC_LOG_OPTS+=("$gc_log_flag:$SOLR_LOGS_DIR/solr_gc.log" '-XX:+UseGCLogFileRotation' '-XX:NumberOfGCLogFiles=9' '-XX:GCLogFileSize=20M')
1802   else
1803     # http://openjdk.java.net/jeps/158
1804     for i in "${!GC_LOG_OPTS[@]}";
1805     do
1806       # for simplicity, we only look at the prefix '-Xlog:gc'
1807       # (if 'all' or multiple tags are used starting with anything other then 'gc' the user is on their own)
1808       # if a single additional ':' exists in param, then there is already an explicit output specifier
1809       GC_LOG_OPTS[$i]=$(echo ${GC_LOG_OPTS[$i]} | sed "s|^\(-Xlog:gc[^:]*$\)|\1:file=$SOLR_LOGS_DIR/solr_gc.log:time,uptime:filecount=9,filesize=20000|")
1810     done
1811   fi
1812 fi
```

Let's just talk about how it works for Java 8 now, as it seems to work as expected for Java 9. The first block of code checks if the user has set GC_LOG_OPTS environment variable. If he hasn't, it proceeds to set the default ones. If he has, it parses the options into an array (line 1791).

In the second block, the confusing part is line 1801. It appends the default options to the options provided by the user. 

So for example, if I pass in ```GC_LOG_OPTS="-XX:NumberOfGCLogFiles=20"```, at the end of line 1801, `GC_LOG_OPTS` will be `('-XX:NumberOfGCLogFiles=20', "$gc_log_flag:$SOLR_LOGS_DIR/solr_gc.log" '-XX:+UseGCLogFileRotation' '-XX:NumberOfGCLogFiles=9' '-XX:GCLogFileSize=20M')`.

Which option will be actually selected is ambiguous and undeterministic. Checking it for GCLogFiles, `ps -ef | grep java | grep GCLogFiles` gave us a `9` which would mean it ignored user input. When we tested it for GCLogFileSize it showed both the values. 

There are a couple of ways to solve the problem as mentioned in the linked JIRA issue.

We chose to go with a new GC_LOG_FILE_OPTS option. The way we've implemented it maintains existing behaviour for those who don't want to change any of their start scripts and provides an option to fix this ambiguous behaviour for those who want to change it. So, the users who want to fix it will now have to provide the GC log file options through the new GC_LOG_FILE_OPTS environment variable. 

The changes are very subtle. We replace line 1801 with,

```bash

1801     if [ -z ${GC_LOG_FILE_OPTS+x} ]; then
1802         GC_LOG_FILE_OPTS=("$gc_log_flag:$SOLR_LOGS_DIR/solr_gc.log" '-XX:+UseGCLogFileRotation' '-XX:NumberOfGCLogFiles=9' '-XX:GCLogFileSize=20M')
1803     else
1804         GC_LOG_FILE_OPTS=($GC_LOG_FILE_OPTS)
1805     fi
1806     GC_LOG_OPTS+=$GC_LOG_FILE_OPTS
```

For Java 9, this ambiguous behavior doesn't seem to be present as the code (line 1809) only sets the default options either when the user hasn't provided any input or any specific file options.

> GC_LOG_OPTS customisation is a little confusing
> -----------------------------------------------
>
>                 Key: SOLR-11233
>                 URL: https://issues.apache.org/jira/browse/SOLR-11233
>             Project: Solr
>          Issue Type: Task
>            Reporter: Christine Poerschke
>            Priority: Minor
>         Attachments: SOLR-11233.patch
>
>
> {{GC_LOG_OPTS}} customisation is currently supported but (pre Java 9) the customised settings are supplemented e.g. https://github.com/apache/lucene-solr/blob/releases/lucene-solr/6.6.0/solr/bin/solr#L1713
> {code}
> GC_LOG_OPTS+=("$gc_log_flag:$SOLR_LOGS_DIR/solr_gc.log" '-XX:+UseGCLogFileRotation' '-XX:NumberOfGCLogFiles=9' '-XX:GCLogFileSize=20M')
> {code}
> This seems unexpected and confusing. Some ideas for making it less confusing:
> * a new {{GC_LOG_FILE_OPTS}} option
> ** the new option can be customised but if unset it would default to existing behaviour
> * use customised GC_LOG_OPTS 'as is'
> ** this would be a change to existing behaviour i.e. the [solr#L1713|https://github.com/apache/lucene-solr/blob/releases/lucene-solr/6.6.0/solr/bin/solr#L1713] settings mentioned above would no longer be appended



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org