However, once cherry-picked the portions to CC, re-running cbcollect (without redaction) ends up with empty files that
MB-47380 added... namely, goxdcr_remote_clusters.json and goxdcr_replications.json.
Spoke with Steve Watanabe, and he pointed me to a few check-ins lately that also touched cbcollect_info. Namely:
Interestingly, when I reverted http://review.couchbase.org/c/ns_server/+/157902 patch locally, the cbcollect works, which means that the newly added XDCR files have content in them.
It also seems like the same situation of 0 byte files in the zip file also occurs to at least one other file. One other example I found used a similar syntax as the new goxdcr files being generated. Specifically, the entry:
(And another non-text file, projector_cprof.log)
Ultimately, without reverting the patch, the 2 goxdcr files + the master_events.log + projector_cprof.log are empty in the zip file. And, reverting the patch seems to flush them just fine (non-0 bytes and with valid data). Thus, I don't think it is specific only to goxdcr portion of the log collection.
Visually, I think the patch looks fine, and atm I don't see anything wrong with it.
What's weird is also the fact that all 3 patches (the 2 mentioned above + goxdcr's portion) are present in master, and cbcollect_info runs just fine in master.
File size was a thought, but there are other small files that were flushed just fine with the changeset.
File descriptor count doesn't seem to be a problem, as ulimit -n is big enough on my machine.