[tor-bugs] #26231 [Internal Services/Service - trac]: Addressing trac performance issues

[Tor Bug Tracker & Wiki]

#26231: Addressing trac performance issues
--------------------------------------------------+-----------------
     Reporter:  hiro                              |      Owner:  qbi
         Type:  defect                            |     Status:  new
     Priority:  Medium                            |  Milestone:
    Component:  Internal Services/Service - trac  |    Version:
     Severity:  Normal                            |   Keywords:
Actual Points:                                    |  Parent ID:
       Points:                                    |   Reviewer:
      Sponsor:                                    |
--------------------------------------------------+-----------------
 We have been having issues with trac for a while now. These are mostly
 caused by a number of crawlers browsing our entire history on trac.

 AFAIK we are not sure if these bots are random crawlers or search engines
 doing their normal duty. We could temporarily look at the log while the
 issue happen and try to see if we can spot known IP addresses or User
 Agent tags.

 Regardless, there are a few things that we could do:

 1. Implement the crawl delay directive:
 https://en.wikipedia.org/wiki/Robots_exclusion_standard#Crawl-
 delay_directive
 There is a good chance this is not going to change anything (especially
 with random bots) but it doesn't affect users and it can be implemented
 right away.

 2. This could be DB related. I.e. the crawler is performing some complex
 sql query and the trac process is crashing because of a timeout or
 something. We know Trac search isn't fast either, so this could be an easy
 explanation.

 3. Could this be a hardware issue? The host seem happy, but maybe we could
 easily upgrade the specs of the machine where trac lives?

 4. Try rate limit on iptables / use something like HA-proxy? This will
 have to be discussed as we would have to change our policy on what we want
 to log. We wouldn't have to store logs for a long time but probably we
 would need to keep some state info / addresses / etc for a few minutes. I
 would consider this a last resort.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/26231>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online