[api.met.no] Srvice disruptions

[api.met.no mailing list]

Dear API users,

Unfortunately we have been experiencing several service difficulties during
the past couple of weeks. Yesterday night one of the server racks fell out,
and all afflicted VMs had to be manually restarted when people got back in
the morning. This meant that api.met.no were running on half capacity for
about six hours.

In addition we have also experienced intermittent slow download times,
usually in the afternoon (CET). Since it seemed this was related to high
traffic volume we've tried to rebalance the load between several servers,
but due to the current Easter systems freeze we can't do this properly with
robust failover until next week.

We are currently investigating the cause of the slow download times.
Possible theories are problems with network routing and/or a DDOS attack
related to the war in Ukraine. While the number of requests haven't
increased and the api backends are mostly idle, we're seeing a sharp rise
in the amount of http connections. Unfortunately there is little we can do
with this until after Easter.

In the mean time, it is important that clients enable gzip compression
(indicated in the Accept request header). Not only will this shorted
transfer time, but since objects are stored in compressed form in the cache
it avoids the extra delay of uncompressing the data before sending to the
client, which is computationally expensive.

We'll get back with more information later. In the mean time, happy easter
time.

Geir Aalberg
API developer
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.met.no/pipermail/api-users/attachments/20220412/59e4e034/attachment.htm>