I am collecting Opentelemetry data from z/OS and Linux, unfortunately the clocks on the two systems are not synchronised.
When I display the data in Jaeger, instead of looking like
Where the transaction took 56.5 milliseconds, the output looks like it took over 26 seconds.
The z/OS data (MQPA and CSQ9) is all squashed to the right because the z/OS clock is out by about 26 seconds.
How to fix it
Within the Jaeger yaml configuration file you can specify
extensions:
jaeger_query:
max_clock_skew_adjust: 30s
This says if it looks like the data has a time skew, and the time is off by less than 30 seconds, then make the timings look sensible.
How does it work?
The Opentelemetry data has information about each span processed. One of the fields is the parent span.
If Jaeger know know that the parent span executed between 13:00:00 and 13:00:01, and the spans from z/OS were for times 13:00:20 to 13:00:21, it looks like the z/OS times are out by 20 seconds, and so Jaeger can compensate for this.
By default max_clock_skew_adjust is 0, so there is no compensation.
