Public instance is very slow

[chriskuehl]

On the public instance:

$ echo asdf | time fpb
[...]
fpb  0.30s user 0.04s system 11% cpu 2.848 total

On a certain company's internal instance:

$ echo asdf | time fpb
[...]
fpb  0.18s user 0.02s system 39% cpu 0.485 total

Something must be misconfigured on the public instance?

(This only applies for new uploads, not for viewing files -- on the public instance, uploaded files are served directly by S3 with CloudFront in front of it, so it's very fast.)

[chriskuehl]

Interestingly, if I run a dev instance of fluffy on the same host as the prod fluffy.cc, it can paste in ~0.6s, whereas the prod one (running in Docker) consistently takes 2+ seconds. Networking misconfiguration maybe?

[chriskuehl]

Upgrading to the latest docker and switching from devicemapper to overlay2 shaved off about a second. I'm now getting roughly ~1.3 seconds to paste a new file, which is closer to what I'd expect, but still a lot slower than hitting dev fluffy...

The other (fast) internal fluffy I compared to also runs in Docker, so something is up...

[chriskuehl]

Added some timing code and compared inside and outside of Docker. Outside of Docker, paste() is ~instant (~0.006s) up until upload_objects is called. That function then takes 0.17s (roughly normal). The entire request (as measured by the flask app) takes ~0.18s.

Inside Docker, there is sometimes a big (~0.16s) delay between objects.append(uf) and when we call upload_objects (between C and D in this diff: https://i.fluffy.cc/H2pkJ5J37NMP8nHrjjvGJQg9MD2DGLm5.html. Sometimes it is "only" ~0.004 seconds. `upload_objects then takes ~0.74s, which is significantly more than outside Docker.