nsq-benchmark-litmus-use-case.txt

# start nsqd:
nsqd -max-msg-size=10024768 -sync-every=1 --mem-queue-size=0 -e2e-processing-latency-percentile=0.90,0.99 -e2e-processing-latency-window-time=10s -statsd-address=localhost:8125 -statsd-interval=10s

# start consumer:
nsq_tail -nsqd-tcp-address=127.0.0.1:4150 -topic=test -channel=nsq_tail > /dev/null

# start producer when ready:
while true; do head -c 600000 /dev/zero|tr '\0' 'A'; echo; done | to_nsq -nsqd-tcp-address=127.0.0.1:4150 --topic=test

## results:
* no errors, no timeouts!
* timeline & measurements:

15:24:00 start nsqd and nsq_tail
15:25:00 start to_nsq.    -> 0      msg
15:27:00 kill nsq_tail    -> 43500  msg
15:29:00 wait             -> 48860  msg, 3380 in nsq_tail disk queue
15:31:00 wait             -> 54380  msg, 6755 in nsq_tail disk queue
15:33:00 restart nsq_tail -> 59592  msg, 9950 in nsq_tail disk queue
15:35:00 wait             -> 81700  msg, 6100 in nsq_tail disk queue
15:37:00 wait             -> 104800 msg, 2500 in nsq_tail disk queue
15:39:00 finish           -> 130000 msg, 0    in nsq_tail disk queue


conclusions:
- in normal mode, 43500 messages in 120s = 362 msg/s (at 200B per metric, is equivalent to 1M metrics per second)
- in degraded mode, 5000 messages in 120s = 41 msg/s (equivalent to 123k metrics per second)
- in recovery mode, 22000 messages in 120s = half of normal mode