800ms intra-region Flycast latency (Amsterdam)

Questions / Help
,
1

(continuing from Fly Postgres Connections to Replicas Slow - #17 by mayailurus)

Does anyone know how I could debug and get rid of an extra 800ms of latency added when going through Flycast? Specifically in Amsterdam. This same issue doesn’t occur in iad or when using .internal addresses.

2 Likes
2

Bumping this hoping someone at Fly will see this after no acknowledgment of the first thread.

3

The mental model that this is automatically AMS → AMS → AMS traffic would be useful to verify, as one small thing…

In other words, the Flycast proxy has nodes or operational loci or such somewhere—but maybe it’s not necessarily the exact same region that the source (Fly-hosted Django) Machine and destination splashcat-db.flycast live in? Perhaps a stale configuration at the middle can result in a “hairpin”:
 

                   AMS   →   ORD   →    AMS
                (django)   (proxy)   (postgres)

              [cat]                     [elephant]

 
splashcat-v-flaps (hairpin route)
 

It’s not clear how to detect or unravel this class of anomalies, though.

(As far as I know, the old standbys like traceroute and mtr don’t work on the fdaa:* lines…)

4

Hi @catgirl, make sure your custom connection string includes the sslmode=disable option, using the default sslmode (prefer) with flycast port 5433 seems to result in a round-trip to the primary instance.

2 Likes
5

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.