[Nix-dev] Random 500 Internal Server Error on cache.nixos.org

[Benno Fünfstück]

Hello list,

my nix-index (https://github.com/bennofs/nix-index) project has the problem
that it seems to almost reliable trigger a 500 internal server error on `
cache.nixos.org`.

Here is the curl log from when that happens:
https://gist.github.com/bennofs/f908afff218099f2dab8bb1f874167ae

Note how the error goes away right as (what I assume is) the cloudfront
cache TTL for the 500 error page expires (5 minutes, note the Age header. I
ran curl every 20 seconds and the `Age` header went up continously starting
from ~`0`, the gist only shows the last 3 runs).

I believe that the bug affects both nar's and narinfos.

At https://forums.aws.amazon.com/message.jspa?messageID=689787, it is said
that:

"It looks like there was an error getting the Metabucket partition, and as
a result the request failed. This is a defined behavior and happens from
time to time.

During normal operations it is not unusual to encounter an occasional error
from S3. When using CloudFront with S3, when a user requests an object,
CloudFront requests the same object from S3 and caches it for future users.
Should S3 encounter an error and is unable to provide the object to
CloudFront at that time it sends the error to CloudFront, which then sends
the error to the end user."

Is the cache perhaps misconfigured, in that it caches 500's as well, making
retries impossible?


Thanks in advance,

Benno Fünfstück
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.science.uu.nl/pipermail/nix-dev/attachments/20170614/08216575/attachment.html>