Pin celery to version 4.4.2 because version 4.4.4 breaks the GWCelery unit tests. (See https://git.ligo.org/emfollow/gwcelery/-/issues/348)
Require ligo-gracedb >= 2.7.5 to take advantage of connection pooling and pick up several bug fixes and regression fixes:
Require gwpy >= 2.0.2 to work around a Matplotlib compatibility bug that was fixed in that version (see https://github.com/gwpy/gwpy/issues/1277).
Close Matplotlib figures that are created during tasks to avoid leaking references and memory.
Adapt to a change in the GraceDB server’s API response for a request to create a label that already exists.
Set the matplotlib backend to
aggin order to fix plot layout glitches that started with matplotlib 3.3.0 when
plt.tight_layoutbecame backend dependent (see https://github.com/matplotlib/matplotlib/pull/15221).
Set FAR threshold for early warning alerts to once per day.
Identify early-warning events using the
EARLY_WARNINGlabel rather than the
EarlyWarningsearch type. The search type is already used to distinguish mock (
MDC) events, so it cannot also be used to indicate early-warning events.
Inhibit GCNs for superevents with the INJ label.
Add configuration variable to disable all but MDC alerts from GCN, and set that variable to True on the production instance.
Skip the preliminary alert timeout for early warning events.
Update the documentation on RAVEN functions and external triggers flow chart.
Change BAYESTAR low frequency cutoff from 30 Hz (the default value) to 15 Hz.
Change playground configuration to read O3 replay data rather than O2 replay data.
Drop dependency on seaborn.
Defer sleekxmpp imports until the VOEvent client starts. This way, sleekxmpp is only imported in the thread that actually uses it. This should speed up worker startup by about 0.1 seconds.
Defer Comet and Twisted imports until they are actually needed by the VOEvent broker. This should speed up worker startup by about 0.2 seconds.
Defer imapclient imports until the email client starts. This should speed up worker startup by about 0.1 seconds.
Improve robustness of the email connection by resetting IMAP IDLE mode at least every 5 minutes and improving error-checking upon disconnection.
Add platform and hostname information to the Flask dashboard.
Ensure external sky maps are normalized.
Fix a bug that was introduced in GWCelery 0.12.1 that broke BAYESTAR localizations for PyCBC events. In 0.12.1, the BAYESTAR data handling was changed to merge together the contents of the coinc.xml and psd.xml.gz files into a single XML document so that BAYESTAR was not sensitive to the order in which the two files were passed to it. PyCBC includes the PSD data in its initial upload, and so its psd.xml.gz file is just a copy of coinc.xml. Merging the two documents together resulted in a single file with two copies of every LIGO-LW table, which broke subsequent parsing.
Fix this by adding a special case for PyCBC to download the coinc.xml file only. This has the nice side effect of reducing the latency for PyCBC events because it is no longer necessary to wait for the additional GraceDB REST API calls involved in uploading and download the additional file.
Skip detchar checks for events which occur in the future.
Delay omegascans until data are available for events in the future.
Enable Zstandard compression of tasks and results to reduce bandwidth into and out of Redis.
Enable receipt confirmation of early warning GCN notices.
If available, use spatial coincidence FAR to determine when to publish a coincident event. Update both time and spatial FAR within superevent when publishable.
Fix bug where the superevent handler can trigger on external events.
Set delay to produce preliminary alert to 0 seconds in the playground configuration. In the production configuration, the delay is still 30 seconds.
Adjust broker transport and worker settings so that the superevent worker respects task priorities. This is seen to reduce the latency of preliminary alerts by about 10 seconds.
gwcelery.tasks.bayestar.localizetask no longer cares about the order in which the
psd.xml.gzfile contents arguments are passed to it because the task now combines the XML documents using
ligolw_add. This allows us to change the immediately upstream task in the localization canvas from an
group. This avoids extra trips of the large file contents blobs into and out of Redis. This is seen to reduce the latency of the localization by about a second.
Produce GCN notices of type
LVC_EARLY_WARNINGfor events that have the
Add a new configuration variable
early_warning_alert_far_thresholdto control the FAR threshold for early warning alerts. In the playground environment, its value is the same as the threshold for ordinary CBC events. In the playground environment, its value is infinity, to generate alerts for all early warning events.
Fix bug where a SubGRBTargeted event would trigger a search in both Fermi and Swift.
GCN_PRELIM_SENTlabel after the GCN notice has been sent. Previously, the
GCN_PRELIM_SENTlabel was added after the GCN notice had been sent and after the GCN Circular template had been created. Since it takes many tens of seconds to create the GCN Circular template, this was distorting latency statistics.
Prioritize processing of
label_addedLVAlert messages over
newLVAlert messages in the superevent manager. The labels
PASTRO_READYmust all be present before we can send a public alert, so processing
label_addedmessages with higher priority may speed up preliminary alerts.
Increase the minimum concurrency of the main GWCelery worker pool from 4 to 8 subprocesses in order to decrease latency.
Append to, and do not overwrite, log files, when starting GWCelery via Condor.
Launch raven coincidence search for sub-threshold GRBs separately for different gamma-ray experiments in order to use different time windows. This enables the joint LVK-Fermi and LVK-Swift targeted searches to be integrated with RAVEN.
Grab subGRB Fermi sky maps from GCN.
Create external sky maps for offline subGRBTargeted Swift uploads.
Document recommended value for the Redis server setting
client-output-buffer-limitin order to prevent disconnection of Celery workers returning large task results. This value was established early in O3, but since it was not in the documentation, we frequently forgot to set it when configuring a Redis server on a new or upgraded system.
Add the unit test for tasks/inference.py.
Upload LALInference DAG files to save the exact commands run for the parameter estimation.
Fix the file names of Bayeswave PSDs.
Capture an exception that is produced when attempting to make an omega scan of data that contains NaNs.
Catch missing trigger_duration when launching check vectors for external events.
Run unit tests for Python 3.8.
Update ligo-followup-advocate to 1.1.6.
Update gracedb-sdk to 0.1.4.
When a GRB or SNEWS GCN is received, upload it to GraceDB with the correct group depending on the value of the VOEvent
role="test", then upload to the
role="observation", then upload to the
Un-pin LALSuite and use the latest stable version (at this time, 6.68).
Do not use Online_PE condor slots for lalinference parameter estimation.
Use Online_PE condor slots for lalinference parameter estimation.
Use Bayeswave PSD for online PE.
Fix a bug in skymap generation with online PE posterior samples.
Reduce the number of bilby runs for test events to less than once per day.
Add systematic error contributions to Fermi-GBM sky maps.
Convert Swift-BAT error radii from 90% C.L. to 1-sigma.
Add INTEGRAL and AGILE MCAL to GRB pipelines.
NOT_GRBto external Fermi candidates that do not meet required threshold of a GRB. This is determined by
Most_Likely_Probquantities supplied with Fermi notices. RAVEN will not consider external events labeled
Automatically generate and upload a graphic showing the source property values by means of a bar chart.
Pin astropy < 4.0 to work around an issue with caching of downloaded data on the Caltech cluster. See https://github.com/astropy/astropy/issues/9970.
Switch from GraceDB REST API calls from gracedb-client to gracedb-sdk to gain increased transaction throughput due to HTTP connection pooling.
vetted=Truekeyword argument for GraceDB API calls to produce VOEvents, because that argument was removed from the GraceDB server and client over a year ago.
Decrease the number of OpenMP workers from 40 to 16, now that gstlal is uploading a reduced number of events.
Add VOEvent broker and receiver configuration for playground environment in order to enable end-to-end testing of transmission to and receipt from GCN.
Fix a bug in the upload of bilby results.
Do not start parameter estimation for mock events uploaded to gracedb.ligo.org.
Calculate joint spatio-temporal FAR automatically for external coincidences. Create the combined skymap when both the GW and external skymaps are available.
Increase the number of retries, with incremental retry backoff, when fetching the Fermi sky map from HEASARC. This is because the Fermi skymap is typically uploaded tens of minutes after the GCN notice from Fermi.
Update to Celery 4.4.0.
Add bullet charts for BAYESTAR coherence-versus-incoherence Bayes factors. The BAYESTAR log Bayes factor for coherence versus incoherence is stored in the FITS file header’s LOGBCI field. For each FITS file that has this header field, make a bullet chart to compare the log Bayes factor to a standard table of threshold confidence levels from Kass & Raftery (1995).
Enable the RAVEN alert pipeline by having the superevent manager listen to the label RAVEN_ALERT.
Use RAVEN VOEvent if RAVEN_ALERT.
Generate emcoinc circular if RAVEN_ALERT instead of EM_COINC.
Increase both CBC and Burst trials factors by one due to enabling the RAVEN pipeline.
gwcelery.tasks.detchar.make_omegascanto reuse GWPy’s own plotting functions, instead of using our own Matplotlib code. This fixes a bug that prevented
make_omegascanfrom working with Astropy 4.0 or later.
Unpin Astropy version, now that
make_omegascanworks with the most recent version.
ADVREQnotification as soon as there is an alert which meets the public alert threshold, regardless of whether its annotations are complete. As a result, follow-up advocates will usually receive notifications about 30 seconds earlier, and will receive notifications even if some of the annotations fail.
Increase the FAR threshold of online PE to the public alert threshold.
Update lalsuite to lalsuite==6.63 and unpin scipy.
Change RAVEN to grab sky map from superevent. Block joint FAR calculation for SNEWS coincidences.
Skip Virgo data when online PE is started on O2Replay data since its statevector cannot be read by gwpy.
Modify RAVEN to run on MDC events.
superevents.processtask to process only complete G events instead of running for all the superevent completeness labels. The behavior for running on the
newtype events remains unchanged.
This is the initial release of GWCelery for O3b.
Changes related to configuration settings
Use the Redis server that is provided by the operating system (e.g. as a systemd unit) rather than starting our own Redis server. This prevents a race condition between the shutdown of Redis and the shutdown of the workers that caused the workers to hang on shutdown.
Update HTCondor accounting group from O2 to O3.
Increase throughput for sky localization tasks by offloading processing of the
openmpCelery queue to 40 workers that are launched via HTCondor on specially configured cluster nodes.
Use mpich as the MPI runtime for parameter estimation.
Use different HTCondor accounting groups for Celery workers depending on whether GWCelery is running in the playground environment (
ligo.dev.o3.cbc.pe.bayestar) or the production environment (
Drop support for Python 3.6 so that we can use the
check_outputkeyword argument that was added to
suprocess.run()in Python 3.7.
Pin gwpy to <= 0.15.0 since the updated gwpy fails to read Virgo’s state vector.
Update ligo-followup-advocate version to 1.1.3.
Changes related to superevent/orchestrator design
Add event completeness to publishability criterion. All three of
EMBRIGHT_READYwill be used to evaluate event completeness for CBC events. Only the
SKYMAP_READYlabel will be used to evaluate completeness for burst events.
EM_Selectedto freeze the preferred event of a superevent and launch a preliminary alert.
Make sub-threshold annotations independent of annotations for superevents which pass public alert threshold.
Prevent second preliminary to be sent in the event of any advocate action. Previously, this was only being prevented for ADVNO.
Make skymaps from parameter estimation public automatically.
Changes related to online parameter estimation
Move a comment attached to posterior samples to the corresponding skymap.
Add bilby online parameter inference workflow.
Fix approximant name used for automatic parameter estimation.
Start parameter estimation on mock events.
Add acceptance tests of parameter estimation.
Use nodes dedicated to online PE also for playground events so that the test runs do not get stuck due to the lack of resources.
Add spins in online PE on playground events so that embright probabilities are calculated based on the posterior samples without errors.
Remove skymap generation from PE DAG so that it will not be generated twice.
Notify which pe pipeline failed for the failure of pe condor jobs.
Changes related to external coincidences
Create RAVEN circular if EM_COINC label is applied to superevent.
Make coincidence FAR synchronous within RAVEN pipeline to fix race condition.
Remove redundant SNEWS handler key.
Remove generation of em_coinc circular when
EM_COINClabel is applied.
Apply EM_COINC to preferred event when coincidence passes RAVEN publishing conditions.
Attempt fetching and uploading Fermi skymap upon receinving GCN notice.
Changes related to skymap generation
Revert back to running BAYESTAR for all
ligo-skymap-from-samplesto speed up skymap generation.
Changes related to automated data quality checks
Create omegascans for all detectors upon creation of new superevent.
check_vectorsupon the creation of a superevent. This will allow subthreshold superevents to be annotated with
Changes to the Flask dashboard
Teach preliminary alert form in Flask dashboard to present a dropdown of events sorted by the preferred event criterion.
Incorporate update circular into flask app.
Add a task to
em_bright.pyto compute and upload source properties upon the upload of
Update ligo-raven version to 1.17.
This is a non-sequential bugfix release based on version 0.8.5.1 to fix the following issue:
Fix a lethal bug in
em_bright.pyintroduced in version 0.8.5. The bug would incorrectly use the snr as the maximum mass of the NS and therefore the source property estimation for pipelines apart from gstlal would be grossly incorrect.
Update ligo-raven version to 1.16.
Fix a bug that prevented retrying of failed GraceDB API calls in the superevent manager.
Add a retry for one more potential GraceDB API failure in the initial and update alert workflows.
In the playground environment only, upload each mock event several times in rapid succession with random jitter in order to simulate multiple pipeline uploads.
Expose events to the public prior to sending any kind of alert: preliminary, initial, update, or retraction. Previously this behavior only occurred for preliminary alerts, which created the unusual and undesirable possibility of a public GCN for an event that is not public. As before, events are only exposed to the public in the production environment, and not in the playground environment.
Propagate sky map file extensions (as in
bayestar.fits.gz,1) to the URLs that are presented in GCN notices.
Generate flattened FITS files and sky map visualizations for all superevents, even those that do not rise to the public alert threshold. Note that as a side effect all superevents will have the
EM_Selectedlabel applied, since it is used as a semaphore to trigger the annotations. The
ADVREQlabel used to serve double duty as the semaphore and also as the wake-up call for follow-up advocates, but now it only serves the latter purpose.
The feature of generating flattened FITS files and sky map plots for all superevents comes as a request from the Fermi and Swift sub-threshold searches.
Delay running BAYESTAR until the superevent’s preferred event has stabilized. BAYESTAR is the most computationally intensive postprocessing task and running it for all events belonging to a superevent was a bottleneck.
For the playground environment only, decrease the timeout for stabilization of the preferred event from 5 minutes to 2 minutes, which is comparable to how long it has taken recent events to settle. This does not affect the configuration of the production environment.
handle_cbc_eventhandler to call gstlal trained ML based inference for source property computation for gstlal triggers.
Apply EM_COINC to superevent and external event in parallel.
Made raven.py tests more robust and have increased coverage.
Removed the feature of p-astro and em_bright reading mean counts, livetimes or ML classifiers from emfollow/data; moved them to lscsoft/p_astro as package data. Added back the
test_tasks_p_astro.pythat was accidentally taken out in
p_astro == 0.8.0.
Fixed a bug in
gracedb.create_tagto handle the scenario when multiple log messages exist for the same filename. The tag is applied to the most recent log message.
Retry GraceDB API calls that fail due to receiving incomplete or malformed HTTP responses, as indicated by
http.client.HTTPExceptionexceptions. This should work around the increased incidence of
RemoteDisconnectedexceptions that we have seen recently, and that caused a delay in sending out the preliminary alert for S190814bv.
Enable Redis integration for Sentry error reporting.
Downgrade lalsuite to 6.54 since
lalinference_pipein 6.59 has a minor bug, which breaks automatic parameter estimation.
Include the number of participating detectors in the preferred event selection criterion for compact binaries: 3-detector events should be preferred over 2-detector events, and 2-detector events over 1-detector events, on the basis of more accurate localization. Ties are still broken by SNR.
SystemExitexceptions from Python command line tools called in Celery tasks and re-raise them as
RuntimeErrorexceptions so that they do not cause the workers to exit.
publictag to data products before sending out an update GCN notice. This will prevent human errors related to not exposing LALInference files before sending a GCN notice.
Don’t read the entire HTTP response from GraceDB POST requests. We only need the HTTP status code. This change might speed up GraceDB API calls a little bit.
Increase preliminary alert timeout back to 5 minutes.
gracedb.add_event_to_supereventidempotent by catching the
HTTPErrorfrom GraceDB that occurs if the superevent has already been created.
Fix bug where neither the space-time nor temporal coincidence far is calculated if external sky map is unavailable.
Update ligo.skymap to 0.1.9. This version changes the data type of the multi-resolution HEALPix format’s
UNIQcolumn from an unsigned integer to a signed integer.
Starting with this version, the Linux builds of ligo.skymap are compiled and optimized using the Intel C Compiler.
Change the trials factor for CBC searches to 4, since SPIIR is performing a single search, and that for burst to 3, since oLIB is not currently in operation.
Downgrade lalsuite to 6.59.
Revert change that tried to fix incorrect key for querying external events. The keys were correct before.
gwcelery.tasks.skymaps.plot_volumetasks a reduced Celery priority as compared to
gwcelery.tasks.bayestar.localizeso that the latter are given preference. This ought to speed up the preparation of preliminary GCN notices because only the latter are required for GCNs but both kinds of tasks compete for slots in the resource-intensive OpenMP queue.
Reduce priority for CBC annotation tasks for events that do not pass the public alert threshold.
Update lalsuite to 6.60.
Ensure gracedb calls to create and update superevents are retried in the event of transient GraceDB API errors.
Update ligo-raven version to 1.15. Apply EM_COINC label in raven.py to give more control and prevent race conditions.
Use the space-time coincidence FAR as the default for RAVEN, use the temporal coincidence FAR when sky maps are not available.
Check if GRB is sub-threshold, set search to be ‘SubGRB’. Pass search through external triggers pipeline and RAVEN.
result_expiressetting from its default value of one day to five minutes. Since we pass large byte strings as task arguments and return values, one day is too long to keep task tombstones in the database. This adjustment should reduce the memory footprint of the Redis server during periods with very high rates of GraceDB uploads.
The downside is that task details will remain browsable in Flower for a much shorter period.
Remove p_astro_gstlal.py module, corresponding test modules, and documentation; p_astro will be reported as a pipeline product from gstlal. The computation of p_astro for all other pipelines is unaffected.
Fix EM_COINC bug where it is being over-applied to superevents.
Fix bug where wrong key was called for querying external events.
The initial alert workflow will now consider only
*.fits.gzsky maps and not
*.fitssky maps for GCN Notices. It was an oversight that we did not exclude
*.fitsfiles from the list of extensions to consider when we updated the handling of multiresolution sky maps.
Catch and retry HTTP 429 (“Too Many Requests”) errors from GraceDB.
Enable Sentry integration for Tornado in order to capture errors from the Flower console.
Fix file extensions for LALInference sky map PNG files: they should be named
Increase the Redis server’s log verbosity in order to help diagnose Redis client connection dropouts.
Run sky map plotting and annotation tasks asynchronously so that they do not block sending preliminary alerts. Their outputs are only for human consumption; they are not needed in order to prepare GCN Notices.
Trigger a preliminary alert for a superevent upon the first time that the preferred event is set to an event that meets the public alert criterion.
This fixes a longstanding issue that has prevented automated preliminary alerts from being sent so far. The preferred event at the instant that the timeout ended did not meet the public alert criterion, but a preferred event that was selected some tens of seconds later did.
Decrease preliminary alert timeout to one minute.
The combined effect of these changes should be to decrease the latency for producing preliminary alerts from 7 minutes to 2 minutes.
Work around a Celery canvas bug that prevented LALInference postprocessing from completing.
Fix a copy-paste error that caused
INJlabels to be ignored when determining whether to send a preliminary alert.
Move RAVEN time coincidence windows to the application configuration.
Document the acceptence tests checklist in the instructions for preparing a release.
Update ligo-raven to version 1.14.
Add a dependency on dnspython to silence the following warning message from SleekXMPP:
DNS: dnspython not found. Can not use SRV lookup.
Pin some recently updated dependencies of Celery that caused unit test failures: amqp <= 2.4.2, kombu <= 4.5.0, vine <= 1.3.0.
Prevent subthreshold GRBs with low reliability from being processed as external events.
Add a task in orchestrator.py to generate FITS files and sky map images automatically whenever an HDF5 posterior samples file is uploaded.
Remove special-case handling of single-instrument events. Now, the eligibility of an event for a public alert is determined only on the basis of its false alarm rate.
Run parameter estimation on nodes dedicted to online-PE.
Emcoinc circular is triggered when RAVEN uploads a coincident FAR.
Pin scipy since scipy>=1.3.0 removes an interpolation function which lalinference postprocessing requires.
Work around a bug in the Sentry Python SDK that caused excessive reporting of certain GraceDB exceptions that are listed in tasks’
autoretry_forsettings. See getsentry/sentry-python#370.
Change the name of BAYESTAR localization files to
bayestar.multiorder.fitsto distinguish them from flat-resolution HEALPix files, which are still named
Reimplement LVAlert listener as a Celery bootstep to avoid needing to track a singleton task using a Redis lock, because Redis locks do not play nicely with Redis persistence. The
--lvalertcommand line option must now be passed in order to enable the LVAlert listener.
Turn on Redis database persistence so that Celery task state is preserved across restarts.
expose_to_publicsetting to disable exposing GraceDB events to the public in all environments except for production.
Update to the latest version of GWPy and un-pin Matplotlib because GWPy now supports Matplotlib 3.1.
Pin LALSuite to version 6.54 because LALInference in LALSuite 6.55 is not compatible with Python 3.
Work around a bug in complex Celery canvases (see celery/celery#5512) that prevented initial GCN notices from being sent. As a side effect of this workaround, the initial, update, and retraction canvases will not automatically expose events to the public.
The preliminary alert canvas still does expose events to the public, so under normal circumstances, the follow-up advocate should not have to manually do that. However, if the event has not been exposed to the public for whatever reason, then the follow-up advocate should expose it to the public manually before applying the
ADVOKlabel. See emfollow/followup-advocate-guide!2.
Reduce the false alarm rate threshold for parameter estimation to decrease cluster load.
Remove redundant LVAlert subscription in handle_lvalert_grb to prevent double calls to RAVEN.
Read template weights for P_astro from hdf5 file using h5py for speedup.
Require matplotlib < 3.1 becuase matplotlib 3.1 breaks importing gwpy:
/usr/local/lib/python3.7/site-packages/gwpy/plot/rc.py:79: in <module> rcParams.get('text.latex.preamble', ) + tex.MACROS), E TypeError: can only concatenate str (not "list") to str
gwcelery.tasks.gracedb.get_eventstake any number of keyword arguments to be passed to corresponding client methods.
Update the superevent
t_0field whenever the preferred event changes.
If the VOEvent broker is disabled by setting
voevent_broker_whitelistto an empty list, then suppress the normal error message that would occur when attempting to send a VOEvent when there are no broker connections.
Rearrange preliminary alert workflow so that sky map plots are generated for the newly added FITS file rather than an older FITS file that coincidentally has the same name.
gwcelery.detchar.check_vectorstask apply all GraceDB log messages in order to increase robustness to recoverable GraceDB API errors.
Port over majority of P_astro code from gwcelery to the p-astro package.
Use cleaned data for parameter estimation.
DQVlabels should be mutually exclusive. When
gwcelery.tasks.detchar.check_vectorsadds one of the
DQVlabels, it will now first remove the other label.
Change exception in VOEevent parsing of Fermi subtreshold alerts to match real incoming alerts.
Update Celery to 4.3.0.
Automatically select the most up-to-date calibration uncertainties for parameter estimation.
orchestrator_timeoutto 300s and the
pe_timeoutto 345s. The previous timeout was not sufficient for the online pipelines to upload all of their possible candidates, hence the extension.
Cycle through llhoft, high latency frames, and low latency frames in detchar’s cache creation.
Add explanations on options in online_pe.jinja2 for those who start parameter estimation based on the ini files uploaded to GraceDB.
Calculate horizon distance with psd.xml.gz to determine the upper limit of distance prior for parameter estimation.
Start parameter estimation when the lowest FAR of the events in a superevent is lower than the threshold.
Update the calibration uncertainties used for parameter estimation.
Handle an exception in VOEvent parsing of Fermi subthreshold alerts due to different param names.
Stop uploading corner plots of intrinsic parameters.
Connect to different GCN servers to receive alerts in the production and playground environments, because GCN does not support multiple receiver connections from the same client IP address to the same server.
Change the preferred event assignment logic to not let accidental candidates like G330298 which have low FAR but high SNR values to become the preferred event. From now on,
superevents.should_publishtakes maximum precedence for selecting the preferred event. The same is also used by orchestrator to expose events.
Update RAVEN coinc FAR task call which uses string params versus un-pickleable class object params.
Make sure to consume the entire response from every GraceDB API request. This will ensure that GraceDB API call has completed before the pipeline continues, and will fix errors like we encountered with S190426c where the pipeline would march along before uploads had finished.
Apply ADVREQ label earlier in the preliminary alert workflow.
Update LALSuite to version 6.54. We are now using a stable version again instead of a nightly build.
Add Nagios checks for GCN connectivity.
Improve uploaded comments so that it is easily understood which event has triggered parameter estimation.
Provide a value for terrestrial count for P_astro for non-gstlal pipelines that is consistent with the FAR threshold used.
Update ligo-followup-advocate to 0.0.28.
Stop using unreviewed cleaned data for parameter estimation.
Update detchar check to analyze full template duration for CBC events.
Fix typo in
gracedb.get_instruments: there was the attribute lookup
single.ifo, which should have been the dictionary lookup
gwcelery.tasks.p_astro_other.choose_snrfor gstlal. This method did not previously expect to be called for gstlal, since it is typically only invoked for other pipelines. However, there is one case when
choose_snris invoked for gstlal, which is when the ranking_data file from gstlal is corrupted with NaNs, causing P_astro for gstlal to fail. Thus, choose_snr has now been fixed to also handle gstlal as a pipeline.
Changed default for em-bright from 2.83 to 3.0 M_sun to be consistent with notices.
Give permissions to read the files under parameter estimation run directories to non-owner people so that rota people can check their progresses. The naming convention of the run directories changed.
EM-Bright ML classification requires review. Until then, give answer based on low-latency estimates.
Compute P_astro with mass-based template weighting. Template weights are now keyed on template parameters, rather than bin numbers. This should make P_astro immune to binning conventions.
Add form to manually send a preliminary GCN Notice.
Fix a typo in
gwcelery.subthat caused the Flower dashboard to fail to start.
Round iDQ p(glitch) to 3 decimal places in GraceDB log message.
Switch log telemetry from the on-premise instance of Sentry at Caltech to a cloud-hosted subscription to sentry.io.
In the playground configuration, the
gwcelery.tasks.gcn.validatetask was producing false alarms because the GCN receiver was receiving VOEvents from the production instance, which would certainly differ in content from VOEvents in the playground instance. Fix this by having
gwcelery.tasks.gcn.validatediscard all VOEvents if the VOEvent broadcaster is disabled.
Update ligo-followup-advocate to 0.0.27.
Wait for 1 minute before parameter estimation in case the preferred event is updated with high latency.
Ensure that P_astro accounts for very loud MBTA and PyCBC events, whose FAR saturate at certain low values depending on instrument combination, but whose SNRs can increase indefinitely.
When a user triggers a Preliminary or Update alert through the Flask interface, create a GraceDB log message to record the username.
The Flask interface will now show a confirmation dialog before sending any alerts.
Add a terrifying warning to the Flask interface to make it clear that the interface is live.
Now that LIGO/Virgo alerts are public, switch the GCN listener that we use to confirm receipt of our own GCN Notices from a managed, private connection to an anonymous, public connection.
Migrate the Flask and Flower dashboards from ldas-jobs.ligo.caltech.edu to emfollow.ligo.caltech.edu. The new URLs are:
Remove the htaccess file from our public_html directory, since the reverse proxy configuration is now the responsibility of system administrators.
Display the GWCelery version number in the Flask application.
Add visualizations for
p_astro.jsonsource classification files.
Calculation of number of instruments is now unified across superevent manager and orchestrator using gracedb method
Enable automated preliminary alerts for all pipelines because disabling them in the orchestrator introduced some issues due to the criteria for releasing a public alert drifting away from the definition of a the preferred event of a superevent. We will instead trust pipelines that are still under review will upload events to the playground rather than the production environment.
Fixed normalization issues with p_astro_gstlal.py; normalization was being applied in the wrong places during Bayes factor computation.
Require celery < 4.3.0 because that version breaks the nagios unit tests.
Update false alarm rate trials factors for preliminary alerts.
Enable sending GCN notices for fully automated preliminary alerts.
Add threshold_snr option in online_pe.jinja2, which is used to determine the upper limit of distance prior.
Use the same criteria to decide whether to expose an event publicly in GraceDB as we use to decide whether to issue a public alert.
Do not issue public alerts for single-instrument GW events.
Disable automated preliminary alerts for all pipelines but gstlal and cWB due to outstanding review items for the other pipelines.
This is the penultimate release before LIGO/Virgo observing run 3 (O3).
Make detchar results easier to read by formatting as HTML table.
Allow iDQ to label DQV onto events based on p(glitch). Adjustable by pipeline.
Move functions in tasks/lalinference.py to lalinference_pipe.py in lalsuite.
Take into account calibration errors in automatic Parameter Estimation.
Do not use margphi option for automatic Parameter Estimation with ROQ waveform since that option is not compatible with ROQ likelihood.
Adjust WSGI middleware configuration to adapt to a change in Werkzeug 0.15.0 that broke redirects on form submission in the Flask app. See https://github.com/pallets/werkzeug/pull/1303.
Use the new
ligo.lwmodule for reading gstlal’s
ranking_data.psd.xml.gzfiles, because these files are now written using the new LIGO-LW format that uses integer row IDs.
Use clean data for parameter estimation.
Use production accounting group for PE runs on gracedb events.
Change threshold from log-likelihood equals 6 to a dynamic threshold that ensures that all gstlal events uploaded to gracedb get assigned a P_astro value.
Fix a bug in translating keys from
source_classification.jsonto keyword arguments for
GraceDB.createVOEventthat caused VOEvents to be missing the
FAR threshold for sending preliminary notices for CBC is changed to 1 per 2 months.
Upload log files when LALInference parameter estimation jobs fail or are aborted.
Changed the filename
Change condor log directory from /var/tmp to ~/.cache/condor since gwcelery workers have separate /var/tmp when they are running as condor jobs and that causes problems when gwcelery tries to read log files.
Limit the maximum version of gwpy to 0.14.0 in order to work around a unit test failure that started with gwpy 0.14.1. See https://git.ligo.org/emfollow/gwcelery/issues/95.
Upload a diff whenever a LIGO/Virgo VOEvent that we receive from GCN does not match the original that we sent.
Wait for low-latency or high-latency frame files being transferred to the cluster before parameter estimation starts.
Fixed exponent in the expression of foreground count in p_astro_other task.
Run the sky map postprocessing and add the
PE_READYtag when LALInference finishes.
EM_COINCtriggered circulars to upload to the superevent page.
p-astro reads mean values from a file on CIT, new mass-gap category added. Removed redundant functions from p_astro_gstlal module.
Continuous deployment on the Caltech cluster now uses a robot keytab and
gsisshinstead of SSH keys and vanilla
sshbecause the new my.ligo.org SSH key management does not support scripted access.
Improve the isolation between the production and playground instances of GWCelery by deploying them under two separate user accounts on the Caltech cluster.
Add functionality for em_bright task to query
emfollow/datafor trained machine learning classifier and report probabilities based on it.
Report an environment tag to Sentry corresponding to the GWCelery configuration module (
development) in order to differentiate log messages from different deployments.
gwcelery condorcommand now identifies jobs that it owns by matching both the job batch name and the working directory. This makes it possible to run multiple isolated instances of GWCelery under HTCondor on the same cluster in different working directories.
Change the conditions for starting parameter estimation. For every CBC superevent, create an
online_pe.inifile suitable for starting LALInference. However, only start LALInference if the false alarm rate is less than once per 2 weeks.
Determine PSD segment length for LALInference automatically based on data availability and data quality.
Add a Flask-based web interface for manually triggering certain tasks such as sending updated GCN notices.
Pass along the GWCelery version number to Sentry.
Upload stdout and stderr when dag creation fails and notifications when submitted job fails in Parameter Estimation
Allow detchar module’s
create_cacheto use gwdatafind when frames are no longer in llhoft.
The Nagios monitoring plugin will now report on the status of LVAlert subscriptions.
Change trials factor to 5 for both CBC and Burst categories. CBC includes the 4 CBC pipelines. Burst includes the 4 searches performed in total by the 2 Burst pipelines. An additional external coincidence search.
Automatically set up PE ini file depending on source parameters reported by detection pipelines.
Fix broken links in log messages due to changes in GraceDB URL routes.
Whenever we send a public VOEvent using GCN, also make the corresponding VOEvent file in GraceDB public.
Don’t include Mollweide projection PNG file in VOEvents. The sky map visualizations take longer to generate than the FITS files themselves, so they were unnecessarily slowing down the preliminary alerts.
Preliminary GCN FAR threshold is modified to be group (CBC, Burst, Test) specific.
Update frame type used in LALInference Parameter Estimation.
Handle cases where
p_astro_gstlal.compute_p_astroreturns NaNs by falling back to
Fix a bug that prevented annotations that are specific to 3D sky maps from being performed for multi-resolution FITS files.
Fetch the graceid for the new event added from the gracedb logs since superevent packet does not provide information as to which event is added in case of type event_added.
Add error handling for nonexistent iDQ frames in detchar module.
Update detchar module configuration for ER13.
This is the release of GWCelery for ER13.
Run two separate instances of Comet, one to act as a broker and one to act as a client. This breaks a cycle that would cause retransmission of GRB notices back to GCN.
Fix a race condition that could cause preliminary alerts to be sent out for events for which data quality checks had failed.
redispackage version because recent updates to Kombu and Billiard seem to have fixed the Nagios unit tests.
Start the Comet VOEvent broker as a subprocess intead of using
multiprocessingand go back to using PyGCN instead of Comet as the VOEvent client. This is a workaround for suspected instability due to a bad interaction between
Reset Matplotlib’s style before running
ligo-skymap-plot-volume. There is some other module (probably in LALSuite) that is messing with the rcparams at module scope, which was causing Mollweide plots to come out with unusual aspect ratios.
check_vectorsupon addition of an event to a superevent if the superevent already has an
Do not check the DMT-DQ_VECTOR for pipelines which use gated h(t).
Remove static example VOEvents from the Open Alert Users Guide. We never used them because activating sample alerts got help until ER13.
Disable running the Orchestrator for test events for ER13. After ER13 is over, we need to carefully audit the code and make sure that test events are handled appropriately.
Enable public GraceDB entries and public GCNs for mock (MDC) events. For real events in ER13, disable public preliminary GCNs. Instead, advocate signoffs will trigger making events and GCN notices public:
ADVOKfor initial notices and
ADVNOfor retraction notices.
Include source classification output (BNS/NSBH/BBH/Terrestrial) in GCN Notices.
redispackage version at <3 because the latest version of redis breaks the Nagios unit tests.
Ditch our own homebrew VOEvent broker and use Comet instead.
In addition to traditional flat, fixed-nside sky maps, BAYESTAR will now also upload an experimental multiresolution format described in LIGO-G1800186-v4.
Update URL for static example event.
Add tasks for submitting HTCondor DAGs.
Add a new module,
gwcelery.tasks.lalinference, which provides tasks to start parameter estimation with LALInference and upload the results to GraceDB.
Depend on lalsuite nightly build from 2018-11-04 to pick up changes to LALInference for Python 3 support.
Send static example VOEvents from the Open Alert Users Guide. This will provide a stream of example alerts for astronomers until GraceDB is ready for public access.
Add trials factor correction to the event FAR when comparing against FAR threshold to send out preliminary GCN.
Require that LIGO/Virgo VOEvents that we receive from GCN match the original VOEvents from GraceDB byte-for-byte, since GCN will now pass through our VOEvents without modification.
Work around a bug in astropy.visualization.wcsaxes that affected all-sky plots when Matplotlib’s
text.usetexrcparam is set to
True(https://github.com/astropy/astropy/issues/8004). This bug has evidently been present since at least astropy 1.3, but was not being triggered until recently: it is likely that some other package that we import (e.g. lalsuite) is now globally setting
A try except is added around updateSuperevent to handle a bad request error from server side when updating superevent parameters which have nearby values.
Send automatic preliminary alerts only for events with a false alarm rate below a maximum value specified by a new configuration variable,
State vector vetoes will not suppress processing of preliminary sky maps and source classification. They will still suppress sending preliminary alerts.
Truefor all automated VOEvents.
Preliminary GCN is not sent for superevents created from offline gw events.
gwcelery.tasks.detchar, which uploads a DQR-compatible json to GraceDB with the results of the detchar checks.
Depend on ligo.skymap >= 0.0.17.
Fix a bug in sending initial, update, and retraction GCN notices: we were sending the VOEvent filenames instead of the file contents.
vettedflag to true for all initial, update, and retraction alerts that are triggered by GraceDB signoffs.
Write GraceDB signoffs, instead of just labels, to simulate initial and retraction alerts for mock events, because merely creating the
ADVOKlabel does not cause GraceDB to erase the
ADVREQlabel. This change makes mock alerts more realistic.
Change filename of cWB sky maps from
cWB.fits.gzfor consistency with other pipelines.
Any time that we send a VOEvent, first change the GraceDB permissions on the corresponding superevent so that it is visible to the public. Note that this has no effect during the ongoing software engineering runs because LVEM and unauthenticated access are currently disabled in GraceDB.
publictag instead of the
lvemtag to mark preliminary sky maps for public access rather than LV-EM partner access. Note that GraceDB has not yet actually implemented unauthenticated access, so this should have no effect during our ongoing software engineering runs.
check_idqfunction to detchar module, which reads probabilities generated by iDQ.
DQVlabels should not trigger retraction notices because they prevent preliminary notices from being sent in the first place.
The criterion for selecting a superevent’s preferred event now prefers multiple-detector events to single-detector events, with precedence over source type (CBC versus burst). Any remaining tie is broken by using SNR for CBC and FAR for Burst triggers.
By default, initial and update alerts will find and send the most recently added public sky map.
The initial and update sky maps no longer perform sky map annotations, because they would only be duplicating the annotations performed as part of the preliminary alert.
Mock events now include example initial and retraction notices. Two minutes after each mock event is uploaded, there will be either an
ADVNOlabel applied at random, triggering either an initial or a retraction notice respectively.
Depend on ligo-gracedb >= 2.0.1 in order to pull in a bug fix for VOEvents with ProbHasNS or ProbHasRemnant set to 0.0.
sentry-sdkpackage instead of the deprecated
ravenpackage for Sentry integration.
Separated the external GCN listening handlers into two: one that listens to GCNs about SNEWS triggers and another that listens to Fermi and Swift.
Fixed calls to the raven temporal coincidence search so that search results separate SNEWS triggers from Fermi and Swift.
Add space-time FAR calculation for GRB and GW superevent coincidences. This only runs when skymaps from both triggers are available to download.
Add human vetting for initial GCN notices. For each new superevent that passes state vector checks, the
ADVREQlabel is applied. Rapid response team users should set their GraceDB notification preferences to alert them on
ADVREQlabels. If a user sets the
ADVOKlabel, then an initial notice is issued. If a user sets the
ADVNOlabel, then a retraction notice is issued.
Update the LVAlert host for gracedb-playground.ligo.org.
Add experimental integration with Sentry for log aggregation and error reporting.
Track API and LVAlert schema changes in ligo-gracedb 2.0.0.
Refactor external trigger handling to separate it from the orchestrator.
Fixed a bug in the VOEvent broker to only issue “iamalive” messages after sending the first VOEvent.
Pass group argument to set time windows appropriately when performing raven coincidence searches. Search in the [-600, 60]s range and [-5, 1]s range around external triggers for Burst events and CBC events respectively. Similarly, search in the [-60, 600]s and [-1, 5]s range around Burst and CBC events for external triggers.
Compute and upload FAR for GRB external trigger/superevent coincidence upon receipt of the EM_COINC label application to a superevent.
Add continuous integration testing for Python 3.7, and run test suite against all supported Python versions (3.6, 3.7).
Update ligo.skymap to 0.0.15.
Manage superevents for production, test, and MDC events separately.
Add some more validation of LIGO/Virgo VOEvents from GCN.
Remove now-unused task
check_vectorsrun for external triggers.
Change the preferred event selection criteria for burst events to be FAR instead of SNR.
gwcelery nagiossubcommand for Nagios monitoring.
Incorporate Virgo DQ veto streams into
Update ligo-raven to 1.3 and ligo-followup-advocate to 0.0.11.
Add a workflow graph to superevents module documentation.
gwcelery condor resubmitas a shortcut for
gwcelery condor rm; gwcelery condor submit.
Fix deprecation warning due to renaming of
Update ligo-gracedb to 2.0.0.dev1.
Add injection checks to
Bitmasks are now defined symbolically in
Refactor configuration so that it is possible to customize settings through an environment variable.
The preferred event for superevents is now decided based on higher SNR value instead of lower FAR in the case of a tie between groups.
A check for the existence of the gstlal trigger database is performed so that compute_p_astro does not return None.
Fix spelling of the label that is applied to events after p_astro finishes, changed from
Run p_astro calculation for mock events.
Overhaul preliminary alert pipeline so that it is mostly feature complete for both CBC and Burst events, and uses a common code path for both types. Sky map annotations now occur for both CBC and Burst localizations.
Switch to using the pre-registered port 8096 for receiving proprietary LIGO/Virgo alerts on emfollow.ligo.caltech.edu. This means that the capability to receive GCNs requires setting up a site configuration in advance with Scott Barthelmey.
Once we switch to sending public alerts exclusively, then we can switch back to using port 8099 for anonymous access, requiring no prior site configuration.
Reintroduce pipeline-dependent pre/post peeks for
check_vectorafter fixing issue where pipeline information was being looked for in the wrong dictionary.
check_vectorchecks all detectors regardless of instruments used, but only appends labels based on active instruments.
Fix a few issues in the GCN broker:
Decrease the frequency of keepalive (“iamalive” in VOEvent Transport Protocol parlance) packets from once a second to once a minute at the request of Scott Barthelmey.
Fix a possible race condition that might have caused queued VOEvents to be thrown away unsent shortly after a scheduled keepalive packet.
Consume and ignore all keepalive and ack packets from the client so that the receive buffer does not overrun.
gstlalpipeline. The copmutation is launched for all cbc_gstlal triggers.
Revert pipeline-dependent pre/post peeks for
check_vectorbecause they introduced a regression: it caused the orchestrator failed without running any annotations.
Add timeout and keepalive messages to GCN broker.
Update ligo-gracedb to 2.0.0.dev0 and ligo.skymap to 0.0.12.
Add superevent duration for gstlal-spiir pipeline.
Fix fallback for determining superevent duration for unknown pipelines.
check_vectorpre/post peeks pipeline dependent.
Process gstlal-spiir events.
Create combined LVC-Fermi skymap in case of coincident triggers and upload to GraceDB superevent page. Also upload the original external trigger sky map to the external trigger GraceDB page.
Generalize conditional processing of complex canvases by replacing the
continue_if_group_is()task with a more general task that can be used like
check_vector_prepostconfiguration variable to control how much padding is added around an event for querying the state vector time series.
This should have the beneficial side effect of fixing some crashes for burst events, for which the bare duration of the superevent segment was less than one sample.
MBTA events in GraceDB leave the
searchfield blank. Work around this in
gwcelery.tasks.detchar.check_vectorswhere we expected the field to be present.
Track change in GraceDB JSON response for VOEvent creation.
After fixing some minor bugs in code that had not yet been tested live, sending VOEvents to GCN now works.
Rewrite the GCN broker so that it does not require a dedicated worker.
Send VOEvents for preliminary alerts to GCN.
Only perform state vector checks for detectors that were online, according to the preferred event.
Exclude mock data challenge events from state vector checks.
Add detector state vector checks to the preliminary alert workflow.
Undo accidental configuration change in last version.
Stop listening for three unnecessary GCN notice types:
Switch to SleekXMPP for the LVAlert client, instead of PyXMPP2. Because SleekXMPP has first-class support for publish-subscribe, the LVAlert listener can now automatically subscribe to all LVAlert nodes for which our code has handlers. Most of the client code now lives in a new external package, sleek-lvalert.
Change superevent threshold and mock event rate to once per hour.
Always upload external triggers to the ‘External’ group.
Add rudimentary burst event workflow to orchestrator: it just generates VOEvents and circulars.
Create a label in GraceDB whenever
Fix typo that was causing a task to fail.
Decrease orchestrator timeout to 15 seconds.
Change FAR threshold for creation of superevents to 1 per day.
Update ligo-followup-advocate to >= 0.0.10. Re-enable automatic generation of GCN circulars.
Add “EM bright” classification. This is rudimentary and based only on the point mass estimates from the search pipeline because some of the EM bright classifier’s dependencies are not yet ready for Python 3.
Added logic to select CBC events as preferred event over Burst. FAR acts as tie breaker when groups for preferred event and new event match.
BAYESTAR now adds GraceDB URLs of events to FITS headers.
Prevent receiving duplicate copies of LVAlert messages by unregistering redundant LVAlert message types.
Update to ligo-followup-advocate >= 0.0.9 to update GCN Circular text for superevents. Unfortunately, circulars are still disabled due to a regression in ligo-gracedb (see https://git.ligo.org/lscsoft/gracedb-client/issues/7).
Upload BAYESTAR sky maps and annotations to superevents.
Create (but do not send) preliminary VOEvents for all superevents. No vetting is performed yet.
Submit handler tasks to Celery as a single group.
Retry GraceDB tasks that raise a
The superevent handler now skips LVAlert messages that do not affect the false alarm rate of an event (e.g. simple log messages).
(Note that the false alarm rate in GraceDB is set by the initial event upload and can be updated by replacing the event; however replacing the event does not produce an LVAlert message at all, so there is no way to intercept it.)
Added a query kwarg to superevents method to reduce latency in fetching the superevents from gracedb.
Refactored getting event information for update type events so that gracedb is polled only once to get the information needed for superevent manager.
set_preferred_eventtask in gracedb.py to
update_supereventto be a full wrapper around the
updateSupereventclient function. Now it can be used to set preferred event and also update superevent time windows.
cwb(extra) attributes, which should be floating point numbers, are present in lvalert packet as strings. Casting them to avoid embarassing TypeErrors.
Reverted back the typecasting of far, gpstime into float. This is fixed in https://git.ligo.org/lscsoft/gracedb/issues/10
t_endvalues are changed to 1 sec interval.
Added ligo-raven to run on external trigger and superevent creation lvalerts to search for coincidences. In case of coincidence, EM_COINC label is applied to the superevent and external trigger page and the external trigger is added to the list of em_events in superevent object dictionary.
libnodes added to superevent handler.
Events are treated as finite segment window, initial superevent creation with preferred event window. Addition of events to superevents may change the superevent window and also the preferred event.
Change default GraceDB server to https://gracedb-playground.ligo.org/ for open public alert challenge.
Update to ligo-gracedb >= 1.29dev1.
get_supereventsand add a new
get_supereventtask that is a trivial wrapper around
Model the time extent of events and superevents using the
Replace GraceDB.get with GraceDB.superevents from the recent dev release of gracedb-client.
Fix possible false positive matches between GCNs for unrelated GRBs by matching on both TrigID (which is generally the mission elapsed time) and mission name.
Add the configuration variable
superevent_far_thresholdto limit the maximum false alarm rate of events that are included in superevents.
LVAlert handlers are now passed the actual alert data structure rather than the JSON text, so handlers are no longer responsible for calling
json.loads. It is a little bit more convenient and possibly also faster for Celery to deserialize the alert messages.
Playgroundapplication configuration objects in order to facilitate quickly switching between GraceDB servers.
Pipeline specific start and end times for superevent segments. These values are controlled via configuration variables.
Add missing LVAlert message types to superevent handler.
Add some logging to the GCN and LVAlert dispatch code in order to diagnose missed messages.
Ingest Swift, Fermi, and SNEWS GCN notices and save them in GraceDB.
Depend on the pre-release version of the GraceDB client, ligo-gracedb 1.29.dev0, because this is the only version that supports superevents at the moment.
Generate GCN Circular drafts using ligo-followup-advocate.
In the continuous integration pipeline, validate PEP8 naming conventions using pep8-naming.
Add instructions for measuring test coverage and running the linter locally to the contributing guide.
gwcelery.tasks.gcnto make it clear that this submodule contains functionality related to GCN notices, rather than VOEvents in general.
gwcelery.tasks.orchestratorto make it clear that this module encapsulates the behavior associated with the “orchestrator” in the O3 low-latency design document.
Mock up calls to BAYESTAR in test suite to speed it up.
Unify dispatch of LVAlert and GCN messages using decorators. GCN notice handlers are declared like this:
import lxml.etree from gwcelery.tasks import gcn @gcn.handler(gcn.NoticeType.FERMI_GBM_GND_POS, gcn.NoticeType.FERMI_GBM_FIN_POS) def handle_fermi(payload): root = lxml.etree.fromstring(payload) # do work here...
LVAlert message handlers are declared like this:
import json from gwcelery.tasks import lvalert @lvalert.handler('cbc_gstlal', 'cbc_pycbc', 'cbc_mbta') def handle_cbc(alert_content): alert = json.loads(alert_content) # do work here...
Instead of carrying around the GraceDB service URL in tasks, store the GraceDB host name in the Celery application config.
Create superevents by simple clustering in time. Currently this is only supported by the
Disable socket access during most unit tests. This adds some extra assurance that we don’t accidentally interact with production servers during the unit tests.
Ignore BAYESTAR jobs that raise a
DetectorDisablederror. These exceptions are used for control flow and do not constitute a real error. Ignoring these jobs avoids polluting logs and the Flower monitor.
FITS history and comment entries are now displayed in a monospaced font.
Adjust error reporting for some tasks.
Depend on newer version of
Add unit tests for the
gwcelery condor submitsubcommand.
Fix some compatibility issues between the
gwcelery condor submitsubcommand and the format of
condor_q -totals -xmlwith older versions of HTCondor.
gwcelery condor submitand related subcommands as shortcuts for managing GWCelery running under HTCondor.
This is the initial release. It provides rapid sky localization with BAYESTAR, sky map annotation, and sending mock alerts.
By default, GWCelery is configured to listen to the test LVAlert server.
Sending VOEvents to GCN/TAN is disabled for now.