fixed issue where remote agents could create duplicates in hub

This commit is contained in:
nick committed 2026-09-22 23:03:51 +00:00
1 parent 1b200dc009
commit 018d7344f4
1 file changed
+56 -6
+56 -6
View File
@@ -4,8 +4,8 @@ version: 1.0.0
# Dozzle log viewer hub.
#
# Aggregates container logs from the dozzle agents running on the other Flatcar
# VMs (see butane/oswl_connector.yaml.tftpl) and, via the local socket, from
# this host's own containers. Reachable at http://192.168.7.63:8080.
# VMs (see the dozzle-agent units in the other butane configs) and, via the
# local socket, from this host's own containers. Reachable on :8080.
#
# No state is kept on disk: /data is a tmpfs, so the container is fully
# disposable and there is nothing to back up. That is also why simple auth
@@ -13,8 +13,9 @@ version: 1.0.0
#
# Hub <-> agent traffic is TLS, using the certificate pair baked into the
# dozzle image. Every hub and agent on the same image version shares it, so
# nothing needs mounting here. To move to a custom pair later, generate one
# with (do not run as root, writes into $PWD):
# nothing needs mounting here -- which is also why the agents must stay pinned
# to the same tag as the hub. To move to a custom pair later, generate one
# with (writes into $PWD):
#
# docker run --rm -v "$PWD":/out amir20/dozzle:v11.1.1 generate-certs \
# --cert-out /out/dozzle_cert.pem --key-out /out/dozzle_key.pem
@@ -34,20 +35,69 @@ systemd:
Requires=docker.service
[Service]
TimeoutStartSec=0
# Kept as unit environment rather than inline -e values so that
# wait-for-agents.sh can read the agent list from the same place the
# container does. docker run forwards them by bare name below.
Environment="DOZZLE_HOSTNAME=Oswald Logs"
Environment="DOZZLE_ENABLE_MCP=true"
Environment="DOZZLE_REMOTE_AGENT=192.168.7.60:7007|oswald-runner|Goose,192.168.7.61:7007|ai|Goose,192.168.7.62:7007|oswl-connector|Goose"
# Pulled up front so a slow first pull is not mistaken for a hung unit.
ExecStartPre=/usr/bin/docker pull amir20/dozzle:v11.1.1
# Leading "-" so a missing container is not a failure.
ExecStartPre=-/usr/bin/docker rm --force dozzle
# If the hub starts before the agents are listening, the failed initial
# handshake leaves a permanently-offline placeholder host in the UI
# alongside the real one once it connects. Bounded and "-" prefixed, so
# an agent that is genuinely gone delays the hub but never blocks it.
ExecStartPre=-/usr/bin/timeout 180 /opt/dozzle/wait-for-agents.sh
# Foreground (no -d) so systemd supervises the container process.
ExecStart=/usr/bin/docker run --rm --name dozzle \
-v /var/run/docker.sock:/var/run/docker.sock:ro \
--tmpfs /data \
-p 8080:8080 \
-e "DOZZLE_HOSTNAME=Oswald Logs" \
-e "DOZZLE_REMOTE_AGENT=192.168.7.60:7007|oswald-runner|Infra,192.168.7.61:7007|ai|Infra,192.168.7.62:7007|oswl-connector|Infra" \
-e DOZZLE_HOSTNAME \
-e DOZZLE_ENABLE_MCP \
-e DOZZLE_REMOTE_AGENT \
amir20/dozzle:v11.1.1
ExecStop=/usr/bin/docker stop dozzle
Restart=always
RestartSec=5s
[Install]
WantedBy=multi-user.target
storage:
files:
# Blocks until every agent named in DOZZLE_REMOTE_AGENT accepts a TCP
# connection. The list is read from the environment, so adding or removing
# an agent above needs no change here.
- path: /opt/dozzle/wait-for-agents.sh
mode: 0755
contents:
inline: |
#!/usr/bin/bash
# Note: /dev/tcp is unavailable -- Flatcar's bash is built without
# net redirections -- so reachability is probed with curl. The agent
# speaks gRPC over TLS, so any completed connection answers with a
# protocol error; only exit 7 (refused) and 28 (timeout) mean the
# agent is not up yet.
if [ -z "$DOZZLE_REMOTE_AGENT" ]; then
exit 0
fi
for entry in $(echo "$DOZZLE_REMOTE_AGENT" | tr ',' ' '); do
# Strip the |name|group suffix; what is left is host:port.
hostport=$(echo "$entry" | cut -d'|' -f1)
while true; do
curl -sk -o /dev/null --max-time 3 "https://$hostport"
rc=$?
case "$rc" in
7|28)
echo "waiting for dozzle agent $hostport"
sleep 2
;;
*)
break
;;
esac
done
done