565478 - Consider implementing back-off and retry capabilities for Test Pilot 1.0 add-on

Status: RESOLVED DUPLICATE of bug 551719

(Mozilla Labs Graveyard :: Test Pilot, defect)

Description

[Daniel Einspanjer [:dre] [:deinspanjer]]

If the cluster is under heavy load, we might want to design the system to respond with an HTTP 503 code.  This response can include a Retry-After header indicating to the client how long they should wait before attempting the submission request again.

If the proxy were not accepting requests at all, the client would likely get a connection failed error.  We should decide what we want it to do in this case.

We should consider whether we want the client to take any action after receiving an HTTP response 413 - Request Entity Too Large.  This should be the response made if the client is attempting to submit too large of a data file to us (as determined by our configuration).

Comment 1

[Jono Xia]

We already implemented random exponential backoff in https://bugzilla.mozilla.org/show_bug.cgi?id=551719 .  Do we still need to do anything else?

Comment 2

[Daniel Einspanjer [:dre] [:deinspanjer]]

Nope.  As long as this code is executed when it receives a non 200 success from the server, this is good enough for 1.0.
Does the client currently differentiate between different types of failures? i.e. Server busy vs access denied or 500 Server Error?
I ask because there are certain types of responses that would indicate the request would never work and in that case, the client should just drop it rather than retrying several times.