Open myllynen opened 6 years ago
@myllynen I'm experimenting with a change to pmlogger that will defer writing the label records and the preamble (which is a synthetic pmResult) until the first real pmResult is seen, then using the timestamp from the first pmResult in the label records and the preamble (so we'll get two pmResults at the start with the same timestamp). This certainly fixes your pmdumptext example, but I need to be sure there are no other unexpected side-effects.
@kmcdonell I have missed your comment here earlier. Did you have a chance to test this further, did you come across any side-effects? Thanks.
As discussed in https://groups.io/g/pcp/topic/6111366, when creating an archive with pmlogger(1) the prologue is logged ~0.02 seconds before the actual data and we see for example with pmdumptext(1) shows results being 10 minutes off and misses some records in the end. I'm providing a copypaste and a test archive to demonstrate this issue below.
test-archive.tar.gz
Let's use this issue to document expectations and track possible fixes around this. Thanks.