Skip to content

Latest commit

 

History

History
100 lines (73 loc) · 3.94 KB

troubleshooting.rst

File metadata and controls

100 lines (73 loc) · 3.94 KB

Troubleshooting

Start by recording the package version and environment inventory, then check declared dependency consistency:

pycwb --version
python --version
python -m pip list
python -m pip check

If PycWB cannot be imported, confirm the active interpreter with python -c "import sys; print(sys.executable)" and reinstall into the environment described by :ref:`installing_pycwb`.

Command not found or unrecognized command

Activate the environment in which PycWB was installed. Try python -m pycwb --help to list the commands available in that environment.

Configuration rejected

pycwb run checks the configuration when it loads your YAML. Use the error message to locate the rejected setting. Render configuration templates first. The detector key is ifo, not ifos. iwindow is the full time window in seconds. For sky masks and distributions, include explicit angle units and the coordinate frame. See :ref:`schema` and :ref:`coordinate_systems_angles`.

Cross-talk catalog unavailable

A first run downloads a missing catalog from the public PycWB cross-talk data repository. On an offline machine, copy a compatible catalog from an online machine and set filter_dir and wdmXTalk explicitly in your YAML. Record the checksum and verify that the catalog covers l_low through l_high.

If the download was interrupted, preserve the error log and replace the incomplete file with a verified copy. Repeatedly starting the same search does not repair a corrupt existing catalog automatically.

Slow first run or memory exhaustion

The first run compiles Numba/JAX kernels and may build cross-talk caches. Compare warm runs separately from cold runs. For the demo, use one worker and its bundled sky resolution. Adding workers can increase memory use. For production, measure a representative segment before choosing worker counts and scheduler memory. See :ref:`dev_performance` and :ref:`tutorial_search` for execution planning.

No recovered events

First check completion:

pycwb progress --work-dir my_first_search --verbose

Inspect the run log, injection time, detector network and search band. For a modified example, a faint injection can legitimately be missed. Do not change production thresholds merely to force a detection. Compare with a fresh, unmodified demo to separate installation problems from analysis choices.

Missing frames or data-quality files

Check that frame lists and data-quality intervals cover the requested GPS range in every detector. Relative paths are interpreted by the configured workflow; run from the intended working directory or provide explicit paths. Public GWOSC data and collaboration-restricted frame services have different access paths.

Interrupted production or existing output

For a tutorial, create a fresh directory so results are unambiguous. Production recovery requires identifying completed job/trial/lag tuples with progress and preserving configuration and batch membership. --jobs, --trial-idx and --lags select work; consult their exact syntax in :ref:`cli_reference`. Do not treat --force-overwrite as a general resume command. Archive existing results and test the intended recovery selection on a copy before merging.

Manifest missing or mismatched

Copy the referenced jobs.parquet together with the catalog and preserve the relative layout. A missing manifest is a provenance error, not an empty job list. Do not remove catalog metadata to suppress it. See :ref:`reproducibility`.

Still stuck

Use :ref:`support` and include the failing command, version, minimal YAML, relevant traceback, and whether the CLI example succeeds.