Skip to content

Monitoring and logs

Log paths are relative to the process’s working directory:

File Contents
logs/revo-server-runtime.log Startup, configuration, registration, server script messages, replication reports and shutdown.
logs/revo-network-runtime.log Transport diagnostics, when emitted.
logs/revo-voice-runtime.log Voice listener, authentication and routing messages. Embedded voice writes in the game server’s directory. External voice writes in the voice process’s directory.

Each file is created when its log category first writes a message.

Follow the game-server log on Windows:

Terminal window
Get-Content .\logs\revo-server-runtime.log -Tail 100 -Wait

On Linux:

Terminal window
tail -n 100 -F logs/revo-server-runtime.log

With the systemd service from Production deployment, use journalctl -u revo-server -f for console output. The Windows service wrapper captures console output in its own log files.

In JavaScript, revo.log(...) writes to the server runtime log with a [script] prefix. Node’s console.log(...) and console.error(...) write to standard output and standard error, which are captured by the service manager. Include operation names and durations in game-mode logs when investigating slow callbacks or database requests.

Automatic interval changes and admission changes appear in the server runtime log when the corresponding settings are enabled:

Message Meaning
Auto sync rate: 25 -> 30ms ... The automatic controller increased the sync interval. A larger interval reduces update frequency.
Overload guard: admissions paused ... New admissions are paused after sustained load. Existing players remain connected.
Overload guard: admissions resumed ... The recovery conditions were met and new admissions are allowed again.
Auto sync bandwidth: ... or Overload guard bandwidth: ... Outgoing game traffic compared with the configured [transport] bandwidth_budget, in bytes per second.
... bandwidth: sample unavailable A valid bandwidth sample could not be collected.

These messages are written when the interval or admission state changes. The tick value is the average time between server-loop samples, including work, configured waiting and scheduling delays. busiest worker is the busiest replication worker’s active-time fraction during that sample.

With sync_interval = "auto", the interval stays between 25 and 45 ms. With overload_guard = true, continued overload can pause admissions. When both are enabled, the guard waits until the interval has reached 45 ms.

The bandwidth figure covers game connections. Monitor voice and resource downloads separately, along with the host’s total network traffic. The configured budget is a target for load protection.

On a test instance with replication workers enabled, set this at the top of config.toml and restart:

dev = true

Reports require a nonzero worker count under [workers].replication. If it is "auto", check the resolved Replication Workers value in the startup log. The server writes reports roughly every five seconds while its loop is running. The same dev setting also marks the server as a development server in the API listing.

Replication workers: contains totals across the workers:

Field Meaning
published Cumulative updates submitted to workers. An update submitted to several workers is counted for each one.
coalesced Cumulative pending worker updates replaced by newer state.
superseded Cumulative pending movement records replaced by newer state for an individual connection.
sent Cumulative sync records accepted by the transport. Several records can share one transport message.
dropped Cumulative updates rejected because a worker’s input queue reached its limit.
failed Cumulative failed sync sends or rejected movement records at a connection’s queue limit.
busy_us Accumulated worker execution time in microseconds, summed across workers.
pending_bytes Current queued worker input and retained movement data. Excludes work already taken out for processing.
peak_bytes Sum of worker input-queue high-water marks. Retained per-connection movement is excluded from this field.
reliable_bytes, reliable_messages Current reliable data waiting for transport acceptance.

Compare cumulative-counter differences between reports. For example, the change in sent divided by elapsed seconds gives the sync-record submission rate. Multiple workers can accumulate execution time simultaneously, so the change in busy_us can exceed the elapsed wall-clock time.

coalesced and superseded track replacement of older queued state with the latest state. Read them alongside delivery behavior and queue trends. Repeated increases in dropped or failed need investigation.

Replication transport: shows up to three connections with the highest estimated queue delay at that sample. The connection handle identifies a transport session and can change when a player reconnects.

Field Meaning
pending_reliable Reliable bytes waiting to be sent, including scheduled retransmissions.
pending_movement Unreliable bytes waiting to be sent.
unacked Reliable bytes sent and awaiting acknowledgement.
queue_us Estimated wait before newly queued data starts sending, in microseconds.
send_rate The transport’s current estimated send rate for this connection, in bytes per second.
out_bytes, in_bytes Recent outgoing and incoming rates, in bytes per second.
local_quality, remote_quality Delivery quality observed locally and by the peer. Values closer to 1 indicate better delivery.

Convert queue_us to milliseconds by dividing by 1,000. For example, 50000 means about 50 ms of queue delay before transmission. Compare repeated samples for the same connection. The three reported connections can change between samples, so their rates cannot be used as the server’s total bandwidth.

Collect a sample before changing intervals, worker counts or send limits:

Observation Next check
Tick time rises while many players experience delays Compare per-core CPU activity with game-mode callback and database timings. Low average host CPU use can still include a saturated main thread.
A worker stays busy and queued work keeps growing Record the connected player count, streamed entity counts and script-event volume. Reproduce the workload on a test instance.
One connection repeatedly has a large queue or low delivery quality Compare that player’s connection with another player on a different network. Check the per-connection send limit and the host’s uplink usage.
Many connections develop queues at once Compare game traffic, voice traffic and host egress with available uplink capacity. Also check CPU activity and send limits.
Voice fails while gameplay continues Check the voice log and the voice deployment checks.
Joining fails during resource downloads Check the helper’s error and the resource host or CDN response. Keep the resource manifest hash with the report.

Record server-process memory and total network traffic over the same period. Compare results under a similar player count and scene. Resource downloads on a separate CDN contribute to that host’s traffic rather than the game server’s.

For an incident report, include the server version, game-mode build, pinned resource hash, relevant configuration, time of the problem and logs from before and after it. Remove credentials, resource keys and player secrets before sharing.