LOG Meeting Minutes 2017-09-19

Minutes:

  • Upstreaming work:

    • As logs are more visible by way of Elastic Search, we're finding a lot of problems (Luke).

    • Issues with MSO conflating the ONAP transaction ID with its own MSO request IDs. 

    • Issues with MSO propagation of contextual information to BPMN execution. 

    • Slow progress due to the usual issues, plus that we're working across multiple ONAP projects, so we don't always have established points of contact, trust, etc. 

  • Wrappers:

    • Reviewed SDC's Logging wrapper. (Luke, but long story short, because see next point).

    • SDC switching to EELF in 18.02, apparently. 

  • Invocation ID: 

    • We backtracked a bit. Need to review existing MDCs to make sure we haven't already got it covered. 

    • (I don't know that there's necessarily anything new to add. It's not obvious that the MDCs define a scheme, and inasmuch as some could be put to that purpose

    • Confirmed that the EELF audit and metric logs are generally 1:1 with entry and exit, which is very good news. See Format. 

  • Format:

    • Pipes vs. Tabs:

      • Historically pipes (or other character sequences) were to sidestep poor escaping support. (That's Luke's argument anyway, and the earlier guidelines stated that pipes and newlines shouldn't be logged). 

      • It's a matter of taste now that we're escaping. 

      • Vote seems to be for pipes, and that's fine. TBC next call. 

    • MDCs:

      • Besides making output reliably machine-readable, the other consequential change in the guidelines was that MDCs be serialized as name-value pairs rather than by name. 

      • This assists indexing, and:

        • Is basically necessary, since indexing can't be expected to know every MDC and its position in a serialized string for every logger for every component. 

        • It makes provider configuration more maintainable. 

      • This seemed uncontentious for most logs, but there are downstream consequences for AT&T. (Maybe with respect to the audit/metric logs, but maybe also more generally).

    • Consistency:

      • Reaffirmed that we want identical or near-identical output from all (currently supported) ONAP logging. 

      • Reaffirmed that we want to support non-Java logging. 

      • Talked about the implications of using Markers for ENTRY, EXIT, etc. and the possibility of making changes to EELF audit and metric logs so that they're emitted.

    • Syslog and streams came up, but only in passing. 



Actions: