Our log server consumes our log messages through kubernetes pods sysout formatted in json and indexes json fields. We need to specify some predefined fields in messages, so that we can track transactions across pods. For one of our pod we use Liberty profile and have issue to configure logging for these needs.
One idea was to use log4j to send customized json message in console. But all message are corrupted by Liberty log system that handles and modifies all logs done in console. I failed to configure Liberty logging parameters (copySystemStreams = false, console log level = NO) for my needs and always have liberty modify my outputs and interleaved non json messages.
To workaround all that I used liberty consoleFormat="json" logging parameter, but this introduced unnecessary fields and also do not allow me to specify my custom fields.
Is it possible to control liberty logging and console ? What is the best way to do my use case with Liberty (and if possible Log4j)
As you mentioned, Liberty has the ability to log to console in JSON format [1]. The two problems you mentioned with that, for your use case, are 1) unnecessary fields, and 2) did not allow you to specify your custom fields.
Regarding unnecessary fields, Liberty has a fixed set of fields in its JSON schema, which you cannot customize. If you find you don't want some of the fields I can think of a few options:
use Logstash.
Some log handling tools, like Logstash, allow you to remove [2] or mutate [3] fields. If you are sending your logs to Logstash you could adjust the JSON to your needs that way.
change the JSON format Liberty sends to stdout using jq.
The default CMD (from the websphere-liberty:kernel Dockerfile) is:
CMD ["/opt/ibm/wlp/bin/server", "run", "defaultServer"]
You can add your own CMD to your Dockerfile to override that as follows (adjust jq command as needed):
CMD /opt/ibm/wlp/bin/server run defaultServer | grep --line-buffered "}" | jq -c '{ibm_datetime, message}'
If your use case also requires sending log4J output to stdout, I would suggest changing the Dockerfile CMD to run a script you add to the image. In that script you would need to tail your log4J log file as follows (this could be combined with the above advice on how to change the CMD to use jq as well)
`tail -F myLog.json &`
`/opt/ibm/wlp/bin/server run defaultServer`[2] https://www.elastic.co/guide/en/logstash/current/plugins-filters-prune.html
[3] https://www.elastic.co/guide/en/logstash/current/plugins-filters-mutate.html
Just in case it helps, I ran into the same issue and the best solution I found was:
server.xml add <logging consoleSource="message,trace" consoleFormat="json" traceSpecification="{package}={level}"/> (swap package and level as required).bootstrap.properties that contains com.ibm.ws.logging.console.format=json.This will give you consistent server and application logging in JSON. A couple of lines at the boot of the server are not json but that was one empty line and a "Launching defaultServer..." line.
I too wanted the JSON structure to be consistent with other containers using Log4j2 so, I followed the advice from dbourne above and add jq to my CMD in my dockerfile to reformat the JSON:
CMD /opt/ol/wlp/bin/server run defaultServer | stdbuf -o0 -i0 -e0 jq -crR '. as $line | try (fromjson | {level: .loglevel, message: .message, loggerName: .module, thread: .ext_thread}) catch $line'
The stdbuf -o0 -i0 -e0 stops pipe ("|") from buffering its output.
This strips out the liberty specific json attributes, which is either good or bad depending on your perspective. I don't need to new values so I don't have a good recommendation for that.
Although the JUL API is not quite as nice as Log4j2 or SLF4j, it's very little code to wrap the JUL API in something closer to Log4j2 E.g. to have varargs rather than an Object[].
OpenLiberty will also dynamically change logging if you edit the server.xml so, it pretty much has all the necessary bits; IMHO.