Retry /whois to allow for propagation/persistence delay

[reivilibre]

Might solve #1180?

Motivated by matrix-org/synapse#12251, which moves the client IP persistence to a different worker than the one that handles /whois, so there is a propagation delay (when it's on the same worker, the information is enriched with unpersisted IPs).

[richvdh]

is there any way for us to tighten this test - for example, let's say that as soon as /whois returns a 200, the body must be correct?

[reivilibre]

is there any way for us to tighten this test - for example, let's say that as soon as /whois returns a 200, the body must be correct?

There is, but I'm not sure how to go about it:

assert_json_list( $value->{sessions}[0]{connections} );
assert_json_keys( $value->{sessions}[0]{connections}[0], qw( ip last_seen user_agent ) );

If {sessions}[0]{connections}[0] exists, but it doesn't contain the right keys, that would be reason to fail immediately.

How can I escape from retry_until_success elegantly? I see it has the ability to specify a callback that tells it whether something should lead to a retry, but I'm not sure how to piece it together. Do you know of any examples that do this kind of thing off the top of your head, that I could have a look at?

[richvdh]

lgtm!