provide a default timeout on http client

[jtarchie]

When using the rqlite CLI, I could not determine an issue of connection. It turns out that the arguments I had provided were not working. The issue was not present until I added a timeout to the HTTP client.

[otoolep]

Thanks for your contribution. Can you show me what you passed to rqlite? Normally it exits immediately if it can't connect.

I do think your change makes sense.

[jtarchie]

Thanks for your contribution. Can you show me what you passed to rqlite? Normally it exits immediately if it can't connect.

I do think your change makes sense.

I assumed when I passed -s https that it would switch to the correct port 443. When I added the timeout, I finally saw the error message of the scheme://hostname:port and realized what was missing. I understand why the port wasn't changed now.

[otoolep]

@jtarchie -- were you running an rqlited process at the time? If so, what host and port was it listening on? When I try to reproduce this issue I get:

rqlite -s https
ERR! Get "https://127.0.0.1:4001//status": http: server gave HTTP response to HTTPS client

and if nothing is running I get:

rqlite -s https 
ERR! Unable to connect to rqlited at https://127.0.0.1:4001 - is it running?

[jtarchie]

Your example above is when it is run locally. The networking configuration for your host machine might account for timeouts, such as firewalls just dropping packets to any port not being listened on.

For example,

 2023-10-03 06:54:50 ⌚  ~/workspace/rqlite
± |main ✓| → nc localhost 8811

 2023-10-03 06:55:09 ⌚  ~/workspace/rqlite
± |main ✓| → nc localhost 8811 -v
nc: connectx to localhost port 8811 (tcp) failed: Connection refused
nc: connectx to localhost port 8811 (tcp) failed: Connection refused

 2023-10-03 06:55:22 ⌚  ~/workspace/rqlite
± |main ✓| → nc google.com 8811 -v
# this just hangs

 2023-10-03 06:57:13 ⌚  ~/workspace/rqlite
± |main ✓| → nc google.com 8811 -v --apple-tcp-timeout 2
nc: connectx to google.com port 8811 (tcp) failed: Operation timed out
# times out after 2 seconds

I was running through GCP, behind a load balancer I don't control. This is a reproducible example if you are so inclined.

[otoolep]

Ack, thanks, I'll merge this (or something very similar) soon.

[jtarchie]

I can make changes if you want. Please let me know.

[adelowo]

Could the timeout be configurable like the insecure option too? e.g 30s or something that can be parsed with time.ParseDuration

[otoolep]

I thought that might be overkill, but the Go docs say this timeout affects all HTTP requests. So if retrieving backup took longer than the timeout (quite possible) it might break that.

So yeah, let's add a timeout option. It would be:

-t 10s

Default to 30s?

[jtarchie]

I used the built-in functionality of the CLI library. It can parse durations from the command line arguments.

[otoolep]

Thanks!