sccache-dist workers connect to scheduler but never appear in --dist-status

Version: sccache 0.14.0 (prebuilt musl binary)

OS:

• Scheduler/client: Debian (192.168.0.12)
• Worker 1: Fedora (192.168.0.57)
• Worker 2: NixOS (192.168.0.93)

Description:

Workers start successfully and log Server connected to scheduler, but sccache --dist-status on the client consistently shows num_servers: 0, num_cpus: 0.

Occasionally one worker (Fedora) shows up transiently as num_servers: 1, num_cpus: 6, but NixOS never appears despite identical configuration and confirmed network connectivity.

Scheduler config (/etc/sccache/scheduler.conf):

public_addr = "192.168.0.12:10600"

[client_auth]
type = "token"
token = "<redacted>"

[server_auth]
type = "DANGEROUSLY_INSECURE"

Worker config (/etc/sccache/server.conf):

cache_dir = "/tmp/toolchains"
public_addr = "192.168.0.93:10501"
scheduler_url = "http://192.168.0.12:10600"

[builder]
type = "overlay"
build_dir = "/tmp/build"
bwrap_path = "/run/wrappers/bin/bwrap"

[scheduler_auth]
type = "DANGEROUSLY_INSECURE"

Worker logs (relevant):

INFO sccache::dist::http::server] Server listening for clients on 192.168.0.93:10501
INFO sccache::dist::http::server] Server connected to scheduler

Client output

{"SchedulerStatus":["http://192.168.0.12:10600/",{"num_servers":0,"num_cpus":0,"in_progress":0}]}

Verified:

• Port 10501 open and reachable from scheduler (nc -zv confirms)
• Port 10600 open and reachable from workers (curl confirms)
• DANGEROUSLY_INSECURE auth used to rule out JWT issues
• bubblewrap installed and working on all workers
• Restarting scheduler + workers in various orders does not resolve

Expected: Workers should appear in scheduler status after connecting.

Actual: num_servers stays 0 despite workers logging a successful connection.

Start with the sccache-dist scheduler registration path and the sccache --dist-status entry point, then compare them with the worker connection logs and the supplied scheduler/server configurations. Reproduce the setup across the Debian, Fedora, and NixOS hosts and trace why a connected worker is absent from status. Done means the cause is identified and workers reliably appear with accurate server and CPU counts.