Endpoints Debugging¶

Last updated 2 May 2023

This guide outlines some common trouble-shooting steps to take if you are encountering issues using ProxyStore Endpoints.

Test a Local Endpoint¶

Consider you configured and started an endpoint as follows:

$ proxystore-endpoint configure myendpoint
INFO: Configured endpoint myendpoint <f4dc841d-377e-4785-8d66-8eade34f63cd>. Start with:
INFO:   $ proxystore-endpoint start myendpoint
$ proxystore-endpoint start myendpoint
INFO: Starting endpoint process as daemon.
INFO: Logs will be written to ~/.local/share/proxystore/myendpoint/log.txt

Check Endpoint Logs¶

Endpoint logs are written to a directory in $XDG_DATA_HOME/proxystore which in this case is ~/.local/share/proxystore/myendpoint (see home_dir() for the full specification).

$ tail -n 1 ~/.local/share/proxystore/myendpoint/log.txt
INFO  (uvicorn.error) :: Uvicorn running on http://127.0.1.1:8766 (Press CTRL+C to quit)

The logs are the first place to check for any potential issues.

Monitor the Endpoint¶

Debug level logging can be enabled when starting the endpoint, and the endpoint can be run directly in the terminal instead of as a daemon process via the --no-detach flag. These two options are helpful for live monitoring the endpoint.

$ proxystore-endpoint --log-level DEBUG start myendpoint --no-detach

Use the Test CLI¶

The proxystore-endpoint CLI provides a test subcommand for testing endpoint commands. See the CLI Reference.

$ proxystore-endpoint test myendpoint exists abcdef
INFO: Object exists: False

As expected, an object with key abcdef does not exist in the store, but we got a valid response so we know the endpoint is running correctly. You can also validate that this request was logged by the endpoint.

Invoke a REST Request¶

Endpoints serve a REST API so curl can be used to check if an endpoint is accessible. Note the proxystore-endpoint test CLI is preferred for debugging. The correct address the endpoint is listening on can be found in the logs.

$ curl http://127.0.0.1:8765/exists?key=abcdef
{"exists": false}

Warning

If the curl command hangs or returns an HTTP error related to being unable to resolve the URL, check if you have an HTTP proxy set. The proxystore-endpoint test command ignores all HTTP proxies and should work in the case where curl does not. Alternatively, try to unset the http_proxy and HTTP_PROXY environment variables when issuing the command.

Test a Remote Endpoint¶

Consider I have an endpoint running on system A with UUID aaaa0259-5a8c-454b-b17d-61f010d874d4 and another on System B with UUID bbbbab4d-c73a-44ee-a316-58ec8857e83a.

Check Relay Server Connections¶

Both endpoints must be connected to the same relay server to form a peer connection. First, check the address value in the [relay] section is present and set to the correct URI string. The endpoint config is found in the config.toml file in the endpoint directory (e.g., ~/.local/share/proxystore/myendpoint/config.toml). Restart your endpoints if you had to change the configuration.

Second, confirm the endpoint connects to the relay server when started by checking the endpoint logs for a line like this.

INFO  (proxystore.p2p.relay_client) :: Established client connection to relay server at ws://localhost:8765 with client uuid=aaaa0259-5a8c-454b-b17d-61f010d874d4 and name=myendpoint

Use the Test CLI¶

The proxystore-endpoint test CLI can be used to establish a peer connection between two endpoints and invoke remote operations. Here, we will request the endpoint on system A (named "myendpoint") to invoke an exists operation on the endpoint on system B.

$ proxystore-endpoint test --remote bbbbab4d-c73a-44ee-a316-58ec8857e83a myendpoint exists abcdef
INFO: Object exists: False

You will get an error if the peer connection fails. For example:

ERROR: Endpoint returned HTTP error code 500. Request to peer bbbbab4d-c73a-44ee-a316-58ec8857e83a failed: ...

If this happens, check the logs for both endpoints for further error messages. Peer requests typically fail for two reasons:

One of the endpoints is not running (e.g., an endpoint crashed) or is not connected to the relay server.
One of the endpoints is behind a symmetric NAT. The NAT traversal techniques used to establish peer-to-peer connections between endpoints are not reliable across symmetric NATs or poorly behaved legacy NATs.

Check Peer-to-Peer Compatibility¶

After ensuring both endpoints are running and connected to the relay server, you can check the NAT compatibility in two ways.

Endpoints will attempt to discover and log the NAT type on startup, so check the logs to see if this could be the reason.

INFO  (proxystore.p2p.nat) :: Checking NAT type. This may take a moment...
INFO  (proxystore.p2p.nat) :: NAT Type:       Full-cone NAT
INFO  (proxystore.p2p.nat) :: External IP:    <IP ADDRESS>
INFO  (proxystore.p2p.nat) :: External Port:  <PORT>
INFO  (proxystore.p2p.nat) :: NAT traversal for peer-to-peer methods (e.g., hole-punching) is likely to work. (NAT traversal does not work reliably across symmetric NATs or poorly behaved legacy NATs.)

Use the proxystore-endpoint check-nat command to discover your NAT type.

$ proxystore-endpoint check-nat
INFO: Checking NAT type. This may take a moment...
INFO: NAT Type:       Full-cone NAT
INFO: External IP:    <IP ADDRESS>
INFO: External Port:  <PORT>
INFO: NAT traversal for peer-to-peer methods (e.g., hole-punching) is likely to work. (NAT traversal does not work reliably across symmetric NATs or poorly behaved legacy NATs.)