Logfile size in quasar servers

[ben-farnham]

Hi All,

I have a problem with logging (using the Boost rotating logfile flavour of logging) with a specific server.

During investigating a problem with this server, I see no other option but to log extremely verbosely, wait for the error to occur then dig around in the resulting giant logfiles. The boost rotating logfile size is hardcoded to 10*5MB files - 50MB total, and due to the verbosity of what I need to log this is insufficient, it only gives me about the last 5 minutes of log output before the oldest logs are deleted.

Short term of course, I locally change the hardcoded value. Longer term, the rotating log file count and size should be made configurable, that raises questions too…

1. Where should the configuration entries go?
   a. In the StandardMetaData.Log element somewhere gets my vote
2. How to homogenize log configuration for UaTrace and boost? (I know ATLAS prefers UaTrace based logging – already configurable via ServerConfig.xml)
   a. UaTrace configuration appears to be file count and log entry count.
   b. Boost logfile configuration (in the C++ code) is by file count and log file size.
   c. To me the boost approach makes sense; presumably system administrators would be happier to know the precise potential log footprint on disk (i.e. size * count), the UaTrace (log entry * count) seems more vague; how many bytes will a log entry be?
3. Should the log disk footprint (logfile count + size) be modifiable at runtime?
   a. I think not, increasing the footprint is probably fine, reducing it could be more complicated – quasar would discard existing logs. One data entry mistake and possibly useful log information is gone. I think it should be specifiable only at startup - in the config.xml.

So, this looks a lot like further support (and questions) for the internal ticket (OPCUA-686). Any thoughts/comments?

Cheers,
Ben