3.7 KiB

Raw Blame History

Set up log rotation

Nimbus logs are written to stdout, and can be redirected to a file. Writing to a file for a long-running process may lead to difficulties when the file grows large. This is typically solved with a log rotator. A log rotator is responsible for switching the written-to file, as well as compressing and removing old logs.

Using `logrotate`

logrotate provides log rotation and compression. The corresponding package will install its Cron hooks (or Systemd timer) -- all you have to do is add a configuration file for Nimbus in /etc/logrotate.d/nimbus-eth2:

/var/log/nimbus-eth2/*.log {
	compress
	missingok
	copytruncate
}

The above assumes you've configured Nimbus to write its logs to /var/log/nimbus-eth2/ (usually by redirecting stdout and stderr from your init script).

copytruncate is required because, when it comes to moving the log file, logrotate's default behaviour requires application support for re-opening that log file at runtime (something which is currently lacking). So, instead of a move, we tell logrotate to do a copy and a truncation of the existing file. A few log lines may be lost in the process.

You can control rotation frequency and the maximum number of log files kept by using the global configuration file, /etc/logrotate.conf:

# rotate daily
daily
# only keep logs from the last 7 days
rotate 7

Using `rotatelogs`

rotatelogs captures stdout logging and redirects it to a file, rotating and compressing on the fly.

It is available on most servers and can be used with Docker, Systemd and manual setups to write rotated logs files.

In particular, when systemd and its accompanying journald log daemon are used, this setup avoids clogging the system log by keeping the Nimbus logs in a separate location.

Compression

rotatelogs works by reading stdin and redirecting it to a file based on a name pattern. Whenever the log is about to be rotated, the application invokes a shell script with the old and new log files. Our aim is to compress the log file to save space. The Nimbus-eth2 repo provides a helper script that does this:

# Create a rotation script for rotatelogs
cat << EOF > rotatelogs-compress.sh
#!/bin/sh

# Helper script for Apache rotatelogs to compress log files on rotation - `$2` contains the old log file name

if [ -f "$2" ]; then
    # "nice" prevents hogging the CPU with this low-priority task
    nice gzip -9 "$2"
fi
EOF

chmod +x rotatelogs-compress.sh

Run

The final step is to redirect logs to rotatelogs using a pipe when starting Nimbus:

build/nimbus_beacon_node \
  --network:holesky \
  --web3-url="$WEB3URL" \
  --data-dir:$DATADIR 2>&1 | rotatelogs -L "$DATADIR/nbc_bn.log" -p "/path/to/rotatelogs-compress.sh" -D -f -c "$DATADIR/log/nbc_bn_%Y%m%d%H%M%S.log" 3600

The options used in this example do the following:

-L nbc_bn.log - symlinks to the latest log file, for use with tail -F
-p "/path/to/rotatelogs-compress.sh" - runs rotatelogs-compress.sh when rotation is about to happen
-D - creates the log directory if needed
-f - opens the log immediately when starting rotatelogs
-c "$DATADIR/log/nbc_bn_%Y%m%d%H%M%S.log" - includes timestamp in log filename
3600 - rotates logs every hour (3600 seconds)

Deleting old logs

rotatelogs will not do this for you, so you'll need a Cron script (or Systemd timer):

# delete log files older than 7 days
find "$DATADIR/log" -name 'nbc_bn_*.log' -mtime +7 -exec rm '{}' \+

3.7 KiB Raw Blame History