Work with .rrd files

The rerun rrd command can inspect and change .rrd files without writing any code. See the CLI reference for all options.

Most commands write to standard output unless you pass -o, so you can chain them with pipes.

Inspect a file

rerun rrd print my_recording.rrd    # what entities and components are in the file
rerun rrd stats my_recording.rrd    # sizes and chunk statistics
rerun rrd verify my_recording.rrd   # check that this version of Rerun can load the file

Migrate an old file

Rerun can load files from version 0.23 and newer, and migrates them when it loads them. To migrate a file on disk to the current version:

rerun rrd migrate my_recording.rrd

This writes the migrated file in place, and keeps the original as my_recording.backup.rrd. Files from before 0.23 can't be migrated.

Remove data

Drop entities or whole timelines:

rerun rrd filter --drop-entity /camera/image --drop-timeline log_tick my_recording.rrd -o smaller.rrd

--drop-entity matches the exact entity path only, not its children. Pass it once per entity.

Split a file

Split a recording into parts along a timeline, either at given times or into a number of equal parts:

rerun rrd split --timeline frame --time 1000 --time 2000 --output-dir ./parts my_recording.rrd
rerun rrd split --timeline frame --num-parts 4 --output-dir ./parts my_recording.rrd

Each part is a separate recording.

Compact a file

Many small chunks make a file slow to load and query. rerun rrd optimize merges them into fewer, larger chunks:

rerun rrd optimize my_recording.rrd -o my_recording_optimized.rrd

The output also has a footer, which the Viewer needs to load large files on demand. See Optimize chunk count and Open recordings larger than RAM.

Merge files

rerun rrd merge writes the contents of several files into one file:

rerun rrd merge part1.rrd part2.rrd -o merged.rrd

Each recording stays a separate recording in the output. Data from files that share the same application ID and recording ID ends up in the same recording.

To combine files with different IDs into a single recording, give them a common application ID and recording ID with rerun rrd route:

rerun rrd route --application-id my_app --recording-id my_recording part1.rrd part2.rrd -o combined.rrd

rerun rrd route does not decode the data, so it is fast, but it doesn't migrate or check the data either.

You can also open several files together without merging them: rerun part1.rrd part2.rrd. Files with the same application ID and recording ID are shown as one recording.

Append to a file, or rotate files

Rerun does not support appending to an existing .rrd file, or rotating files from the SDK. rr.save() and rr.FileSink overwrite the file if it already exists.

Workarounds:

Write a new file for each session or time span, with the same application ID and recording ID (the first argument and the recording_id argument of rr.init). Open them together, or combine them with rerun rrd merge. See Shared recordings.
To rotate files, stop logging to one file and start a new one with rr.save() (or rr.set_sinks) on the same recording.

Watch a file while it is written

The Viewer does not follow (tail) a file that is still being written. To watch the data live and also save it, send it to both the Viewer and a file:

rr.set_sinks(rr.GrpcSink(), rr.FileSink("my_recording.rrd"))

See Sinks.

Alternatively, stream to standard output and pipe it to the Viewer, which reads from standard input:

python my_script.py | rerun -

Your script needs to call rr.stdout() for this.