Critical - P2 Following measurements were observed with mongod under heavy load using WT:
localhost:27017 357519 *0 *0 *0 0 3431|0 1.5 1.9 0 13.1G 1.8G 0|9509 1|128 12m 395k 10568 07:20:45
localhost:27017 193468 *0 *0 *0 0 2017|0 1.3 1.8 0 13.1G 1.8G 0|9975 2|124 6m 260k 10577 07:20:48
localhost:27017 193468 *0 *0 *0 0 2017|0 1.3 1.8 0 13.1G 1.8G 0|9975 2|124 6m 260k 10577 07:20:48
localhost:27017 214120 *0 *0 *0 0 2276|0 1.3 1.8 0 13.1G 1.8G 0|10079 1|123 7m 285k 10587 07:20:50
insert query update delete getmore command % dirty % used flushes vsize res qr|qw ar|aw netIn netOut conn time
localhost:27017 214120 *0 *0 *0 0 2276|0 1.3 1.8 0 13.1G 1.8G 0|10079 1|123 7m 285k 10587 07:20:50
localhost:27017 214120 *0 *0 *0 0 2276|0 1.3 1.8 0 13.1G 1.8G 0|10079 1|123 7m 285k 10587 07:20:50
localhost:27017 331873 *0 *0 *0 0 3327|0 1.4 1.9 0 13.1G 1.8G 0|9803 1|128 11m 384k 10597 07:20:53
localhost:27017 331873 *0 *0 *0 0 3327|0 1.4 1.9 0 13.1G 1.8G 0|9803 1|128 11m 384k 10597 07:20:53
localhost:27017 109608 *0 *0 *0 0 1014|0 1.4 2.0 0 13.1G 1.8G 0|9620 1|128 4m 148k 10598 07:20:55
localhost:27017 139772 *0 *0 *0 0 1364|0 1.3 2.0 0 13.2G 1.8G 0|9436 2|128 5m 165k 10607 07:20:57
localhost:27017 139772 *0 *0 *0 0 1364|0 1.3 2.0 0 13.2G 1.8G 0|9436 2|128 5m 165k 10607 07:20:57
localhost:27017 139772 *0 *0 *0 0 1364|0 1.3 2.0 0 13.2G 1.8G 0|9436 2|128 5m 165k 10607 07:20:57
localhost:27017 219708 *0 *0 *0 0 2199|0 1.2 1.9 0 13.2G 1.8G 0|9702 1|128 7m 294k 10617 07:21:00
localhost:27017 219708 *0 *0 *0 0 2199|0 1.2 1.9 0 13.2G 1.8G 0|9702 1|128 7m 294k 10617 07:21:00
insert query update delete getmore command % dirty % used flushes vsize res qr|qw ar|aw netIn netOut conn time
localhost:27017 219708 *0 *0 *0 0 2199|0 1.2 1.9 0 13.2G 1.8G 0|9702 1|128 7m 294k 10617 07:21:00
Note several groups of identical lines reporting identical timestamps and identical throughput, with gaps of between the timestamps, which should be at 1-second intervals. I know from independent throughput measurement that the actual throughput during this time period was much lower than that reported above, by a factor of 2-3x. So it appears that
- when serverStatus is slow mongostat repeats the last sample obtained
- the throughput calculation looks like maybe it's based on the expected 1s between serverStatus samples, not on the actual time between samples?