HAProxy for API calls ?

[immo]

We do have a lot of API calls from different external applications.
Each currently pointing to a master host which we expect to have as stable as possible.
Should we put a haproxy beforehand to have to the api calls distributed over the different hosts ?
If the application would be able to get the IPs of the different running nodes we could randomize the api call targets inside the application. But how to get the iplist to do the proxiing by myself. (ok a nslookup could be done for every host but a api return value would be nicer....)

Any proposals for haproxy (first choice ) or the list of node ips.

 

[Stoiko Ivanov]

Should we put a haproxy beforehand to have to the api calls distributed over the different hosts ?

Where would the haproxy run - if it's a single instance this becomes your new single point of failure - no api-requests get routed if the node/host where it's running is down - and setting it up with redundancy and floating IP's is a bit more involved (but can work of course)

As for getting the list of ip's of nodes in a cluster - the api route /cluster/status should contain this information

Hope this helps!

 

[immo]

sure that HAproxy would be a single point of failure too but to widespread the amount of api calls over the different nodes this could help. and for HA proxy redundancy there are also some solutions available

pvesh get /cluster/status
returns just
┌─────────┐
│ type │
├─────────┤
│ cluster │
├─────────┤
│ node │
├─────────┤
│ node │
├─────────┤
│ node │
└─────────┘

 

[Stoiko Ivanov]

try `pvesh get /cluster/status --output-format json-pretty`

(the tabular output of pvesh does not always return all values, and you'd get json-output via the REST-API anyways)