use tokio::time::timeout to timeout fetch state and send encrypted #889
Conversation
| url | ||
| ); | ||
| continue; | ||
| let work = self.http.get(url.clone()).send(); |
There was a problem hiding this comment.
Since we have issues with node stability, should we add this info to metrics? Each node can report the number of active participants. We can even add an alert on this.
There was a problem hiding this comment.
sounds good! I'll do it in another PR since this PR will already be big
ppca
changed the title
ppca
changed the title
ppca
force-pushed
the
xiangyi/tune_refresh_intervals
branch
from
6201a06 to
58d7056
Compare
|
as discussed with @volovyks , since we might replace the client later with sockets etc, we will not refactor the timeouts into contract. |
I see, then let's not bake these sort of configuration into terraform at all and just keep with the defaults in the CLI itself for now |
| @@ -63,6 +63,10 @@ pub enum Cli { | |||
| /// referer header for mainnet whitelist | |||
| #[arg(long, env("MPC_CLIENT_HEADER_REFERER"), default_value(None))] | |||
| client_header_referer: Option<String>, | |||
| #[clap(flatten)] | |||
| mesh_options: mesh::Options, | |||
There was a problem hiding this comment.
Do partners need to update their terraform setup?
The previous implementation caused troubles: 1) 1s timeout have already stuck the protocol; 2) 500ms timeout still see many timeouts on /state endpoint.
I am guessing the .timeout() that came with reqwest package cannot successfully time out an async call, the counting of time may not have considered task/thread switching.
So I switched to using tokio::time::timeout and now problem solved.
Dev success increased from 70% to 100%.
This change sets the timeout to 1000ms for both fetch /state and for send encrypted. Both can be tuned as an environment variable.