I'm working on upgrading a Rails 4.2 application from jruby-1.7.23 to jruby-9.0.4.0. This is a fairly large app, running on trinidad and using MySQL and MongoDB.

I am opening this ticket as suggested by @headius on IRC (Thanks to @enebo too for helping out).

The first indication of performance issues is our specs take way longer. On 1.7.23 it takes 17:55, on 9.0.4.0 it takes 41:15. Over 4K specs running.

Our unmodified code, running locally on trinidad and hitting a single endpoint (it's a pretty slow one to start with, that involves quite a bit of database activity). The benchmark was conducted with siege, twice on each version (for warmup):

$ siege -c 4 -t30s $URL

For 1.7.23:

Transactions:                 59 hits
Availability:             100.00 %
Elapsed time:              29.05 secs
Data transferred:           0.27 MB
Response time:              1.38 secs
Transaction rate:           2.03 trans/sec
Throughput:             0.01 MB/sec
Concurrency:                2.79
Successful transactions:          59
Failed transactions:               0
Longest transaction:            1.92
Shortest transaction:           0.46

For 9.0.4.0

Transactions:                 45 hits
Availability:             100.00 %
Elapsed time:              29.08 secs
Data transferred:           0.20 MB
Response time:              1.96 secs
Transaction rate:           1.55 trans/sec
Throughput:             0.01 MB/sec
Concurrency:                3.03
Successful transactions:          45
Failed transactions:               0
Longest transaction:            3.32
Shortest transaction:           0.60

As you can see the response time is 42% slower.

If I modify the code so that the controller does nothing (except for before_filters and auth token validation) and returns head :ok, then the performance is quite similar (9.0.4.0 is 6% slower). This would indicate that there is something inside our code.

As suggested by @headius, I tried runnig the same benchmark with profiling in --sample mode. You can see the results for both 1.7.23 and 9.0.4.0 in this gist. I would appreciate some help reading the output: I can't tell from that where I should focus my attention on.

I also profiled the inside of the controller action in 9.0.4.0, but I can't see an obvious bottleneck either. The results are in the same gist.

I appreciate guidance and direction to get to the bottom of this.

Thanks,