Coverage for python/lsst/summit/utils/tmaUtils.py: 18%
651 statements
« prev ^ index » next coverage.py v7.4.4, created at 2024-03-28 05:08 -0700
« prev ^ index » next coverage.py v7.4.4, created at 2024-03-28 05:08 -0700
1# This file is part of summit_utils.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (https://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program. If not, see <https://www.gnu.org/licenses/>.
22import datetime
23import enum
24import itertools
25import logging
26import re
27from dataclasses import dataclass, field
29import humanize
30import matplotlib.dates as mdates
31import matplotlib.pyplot as plt
32import numpy as np
33import pandas as pd
34from astropy.time import Time
35from matplotlib.ticker import FuncFormatter
37from lsst.utils.iteration import ensure_iterable
39from .blockUtils import BlockParser
40from .efdUtils import (
41 COMMAND_ALIASES,
42 clipDataToEvent,
43 efdTimestampToAstropy,
44 getCommands,
45 getDayObsEndTime,
46 getDayObsForTime,
47 getDayObsStartTime,
48 getEfdData,
49 makeEfdClient,
50)
51from .enums import AxisMotionState, PowerState
52from .utils import dayObsIntToString, getCurrentDayObs_int
54__all__ = (
55 "TMAStateMachine",
56 "TMAEvent",
57 "TMAEventMaker",
58 "TMAState",
59 "AxisMotionState",
60 "PowerState",
61 "getSlewsFromEventList",
62 "getTracksFromEventList",
63 "getTorqueMaxima",
64 "filterBadValues",
65)
67# we don't want to use `None` for a no data sentinel because dict.get('key')
68# returns None if the key isn't present, and also we need to mark that the data
69# was queried for and no data was found, whereas the key not being present
70# means that we've not yet looked for the data.
71NO_DATA_SENTINEL = "NODATA"
73# The known time difference between the TMA demand position and the TMA
74# position when tracking. 20Hz data times three points = 150ms.
75TRACKING_RESIDUAL_TAIL_CLIP = -0.15 # seconds
78def getSlewsFromEventList(events):
79 """Get the slew events from a list of TMAEvents.
81 Parameters
82 ----------
83 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
84 The list of events to filter.
86 Returns
87 -------
88 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
89 The filtered list of events.
90 """
91 return [e for e in events if e.type == TMAState.SLEWING]
94def getTracksFromEventList(events):
95 """Get the tracking events from a list of TMAEvents.
97 Parameters
98 ----------
99 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
100 The list of events to filter.
102 Returns
103 -------
104 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
105 The filtered list of events.
106 """
107 return [e for e in events if e.type == TMAState.TRACKING]
110def getTorqueMaxima(table):
111 """Print the maximum positive and negative azimuth and elevation torques.
113 Designed to be used with the table as downloaded from RubinTV.
115 Parameters
116 ----------
117 table : `pd.DataFrame`
118 The table of data to use, as generated by Rapid Analysis.
119 """
120 for axis in ["elevation", "azimuth"]:
121 col = f"Largest {axis} torque"
122 maxPos = np.argmax(table[col])
123 maxVal = table[col].iloc[maxPos]
124 print(f"Max positive {axis:9} torque during seqNum {maxPos:>4}: {maxVal/1000:>7.1f}kNm")
125 minPos = np.argmin(table[col])
126 minVal = table[col].iloc[minPos]
127 print(f"Max negative {axis:9} torque during seqNum {minPos:>4}: {minVal/1000:>7.1f}kNm")
130def getAzimuthElevationDataForEvent(
131 client,
132 event,
133 prePadding=0,
134 postPadding=0,
135):
136 """Get the data for the az/el telemetry topics for a given TMAEvent.
138 The error between the actual and demanded positions is calculated and added
139 to the dataframes in the az/elError columns. For TRACKING type events, this
140 error should be extremely close to zero, whereas for SLEWING type events,
141 this error represents the how far the TMA is from the demanded position,
142 and is therefore arbitrarily large, and tends to zero as the TMA get closer
143 to tracking the sky.
145 Parameters
146 ----------
147 client : `lsst_efd_client.efd_helper.EfdClient`
148 The EFD client to use.
149 event : `lsst.summit.utils.tmaUtils.TMAEvent`
150 The event to get the data for.
151 prePadding : `float`, optional
152 The amount of time to pad the event with before the start time, in
153 seconds.
154 postPadding : `float`, optional
155 The amount of time to pad the event with after the end time, in
156 seconds.
158 Returns
159 -------
160 azimuthData : `pd.DataFrame`
161 The azimuth data for the specified event.
162 elevationData : `pd.DataFrame`
163 The elevation data for the specified event.
164 """
165 azimuthData = getEfdData(
166 client, "lsst.sal.MTMount.azimuth", event=event, prePadding=prePadding, postPadding=postPadding
167 )
168 elevationData = getEfdData(
169 client, "lsst.sal.MTMount.elevation", event=event, prePadding=prePadding, postPadding=postPadding
170 )
172 azValues = azimuthData["actualPosition"].values
173 elValues = elevationData["actualPosition"].values
174 azDemand = azimuthData["demandPosition"].values
175 elDemand = elevationData["demandPosition"].values
177 azError = (azValues - azDemand) * 3600
178 elError = (elValues - elDemand) * 3600
180 azimuthData["azError"] = azError
181 elevationData["elError"] = elError
183 return azimuthData, elevationData
186def filterBadValues(values, maxDelta=0.1, maxConsecutiveValues=3):
187 """Filter out bad values from a dataset, replacing them in-place.
189 This function replaces non-physical points in the dataset with an
190 extrapolation of the preceding two values. No more than 3 successive data
191 points are allowed to be replaced. Minimum length of the input is 3 points.
193 Parameters
194 ----------
195 values : `list` or `np.ndarray`
196 The dataset containing the values to be filtered.
197 maxDelta : `float`, optional
198 The maximum allowed difference between consecutive values. Values with
199 a difference greater than `maxDelta` will be considered as bad values
200 and replaced with an extrapolation.
201 maxConsecutiveValues : `int`, optional
202 The maximum number of consecutive values to replace. Defaults to 3.
204 Returns
205 -------
206 nBadPoints : `int`
207 The number of bad values that were replaced out.
208 """
209 # Find non-physical points and replace with extrapolation. No more than
210 # maxConsecutiveValues successive data points can be replaced.
211 badCounter = 0
212 consecutiveCounter = 0
214 log = logging.getLogger(__name__)
216 median = np.nanmedian(values)
217 # if either of the the first two points are more than maxDelta away from
218 # the median, replace them with the median
219 for i in range(2):
220 if abs(values[i] - median) > maxDelta:
221 log.warning(f"Replacing bad value of {values[i]} at index {i} with {median=}")
222 values[i] = median
223 badCounter += 1
225 # from the second element of the array, walk through and calculate the
226 # difference between each element and the previous one. If the difference
227 # is greater than maxDelta, replace the element with the average of the
228 # previous two known good values, i.e. ones which have not been replaced.
229 # if the first two points differ from the median by more than maxDelta,
230 # replace them with the median
231 lastGoodValue1 = values[1] # the most recent good value
232 lastGoodValue2 = values[0] # the second most recent good value
233 replacementValue = (lastGoodValue1 + lastGoodValue2) / 2.0 # in case we have to replace the first value
234 for i in range(2, len(values)):
235 if abs(values[i] - lastGoodValue1) >= maxDelta:
236 if consecutiveCounter < maxConsecutiveValues:
237 consecutiveCounter += 1
238 badCounter += 1
239 log.warning(f"Replacing value at index {i} with {replacementValue}")
240 values[i] = replacementValue
241 else:
242 log.warning(
243 f"More than 3 consecutive replacements at index {i}. Stopping replacements"
244 " until the next good value."
245 )
246 else:
247 lastGoodValue2 = lastGoodValue1
248 lastGoodValue1 = values[i]
249 replacementValue = (lastGoodValue1 + lastGoodValue2) / 2.0
250 consecutiveCounter = 0
251 return badCounter
254def plotEvent(
255 client,
256 event,
257 fig=None,
258 prePadding=0,
259 postPadding=0,
260 commands={},
261 azimuthData=None,
262 elevationData=None,
263 doFilterResiduals=False,
264 maxDelta=0.1,
265):
266 """Plot the TMA axis positions over the course of a given TMAEvent.
268 Plots the axis motion profiles for the given event, with optional padding
269 at the start and end of the event. If the data is provided via the
270 azimuthData and elevationData parameters, it will be used, otherwise it
271 will be queried from the EFD.
273 Optionally plots any commands issued during or around the event, if these
274 are supplied. Commands are supplied as a dictionary of the command topic
275 strings, with values as astro.time.Time objects at which the command was
276 issued.
278 Due to a problem with the way the data is uploaded to the EFD, there are
279 occasional points in the tracking error plots that are very much larger
280 than the typical mount jitter. These points are unphysical, since it is not
281 possible for the mount to move that fast. We don't want these points, which
282 are not true mount problems, to distract from any real mount problems, and
283 these can be filtered out via the ``doFilterResiduals`` kwarg, which
284 replaces these non-physical points with an extrapolation of the average of
285 the preceding two known-good points. If the first two points are bad these
286 are replaced with the median of the dataset. The maximum difference between
287 the model and the actual data, in arcseconds, to allow before filtering a
288 data point can be set with the ``maxDelta`` kwarg.
290 Parameters
291 ----------
292 client : `lsst_efd_client.efd_helper.EfdClient`
293 The EFD client to use.
294 event : `lsst.summit.utils.tmaUtils.TMAEvent`
295 The event to plot.
296 fig : `matplotlib.figure.Figure`, optional
297 The figure to plot on. If not specified, a new figure will be created.
298 prePadding : `float`, optional
299 The amount of time to pad the event with before the start time, in
300 seconds.
301 postPadding : `float`, optional
302 The amount of time to pad the event with after the end time, in
303 seconds.
304 commands : `dict` [`pd.Timestamp`, `str`], or
305 `dict` [`datetime.datetime`, `str`], oroptional
306 A dictionary of commands to plot on the figure. The keys are the times
307 at which a command was issued, and the value is the command string, as
308 returned by efdUtils.getCommands().
309 azimuthData : `pd.DataFrame`, optional
310 The azimuth data to plot. If not specified, it will be queried from the
311 EFD.
312 elevationData : `pd.DataFrame`, optional
313 The elevation data to plot. If not specified, it will be queried from
314 the EFD.
315 doFilterResiduals : 'bool', optional
316 Enables filtering of unphysical data points in the tracking residuals.
317 maxDelta : `float`, optional
318 The maximum difference between the model and the actual data, in
319 arcseconds, to allow before filtering the data point. Ignored if
320 ``doFilterResiduals`` is `False`.
321 Returns
322 -------
323 fig : `matplotlib.figure.Figure`
324 The figure on which the plot was made.
325 """
327 def tickFormatter(value, tick_number):
328 # Convert the value to a string without subtracting large numbers
329 # tick_number is unused.
330 return f"{value:.2f}"
332 def getPlotTime(time):
333 """Get the right time to plot a point from the various time formats."""
334 match time:
335 case pd.Timestamp():
336 return time.to_pydatetime()
337 case Time():
338 return time.utc.datetime
339 case datetime.datetime():
340 return time
341 case _:
342 raise ValueError(f"Unknown type for commandTime: {type(time)}")
344 # plot any commands we might have
345 if not isinstance(commands, dict):
346 raise TypeError("commands must be a dict of command names with values as" " astropy.time.Time values")
348 if fig is None:
349 fig = plt.figure(figsize=(10, 8))
350 log = logging.getLogger(__name__)
351 log.warning(
352 "Making new matplotlib figure - if this is in a loop you're going to have a bad time."
353 " Pass in a figure with fig = plt.figure(figsize=(10, 8)) to avoid this warning."
354 )
356 fig.clear()
357 ax1p5 = None # need to always be defined
358 if event.type.name == "TRACKING":
359 ax1, ax1p5, ax2 = fig.subplots(
360 3, sharex=True, gridspec_kw={"wspace": 0, "hspace": 0, "height_ratios": [2.5, 1, 1]}
361 )
362 else:
363 ax1, ax2 = fig.subplots(
364 2, sharex=True, gridspec_kw={"wspace": 0, "hspace": 0, "height_ratios": [2.5, 1]}
365 )
367 if azimuthData is None or elevationData is None:
368 azimuthData, elevationData = getAzimuthElevationDataForEvent(
369 client, event, prePadding=prePadding, postPadding=postPadding
370 )
372 # Use the native color cycle for the lines. Because they're on different
373 # axes they don't cycle by themselves
374 lineColors = [p["color"] for p in plt.rcParams["axes.prop_cycle"]]
375 nColors = len(lineColors)
376 colorCounter = 0
378 ax1.plot(azimuthData["actualPosition"], label="Azimuth position", c=lineColors[colorCounter % nColors])
379 colorCounter += 1
380 ax1.yaxis.set_major_formatter(FuncFormatter(tickFormatter))
381 ax1.set_ylabel("Azimuth (degrees)")
383 ax1_twin = ax1.twinx()
384 ax1_twin.plot(
385 elevationData["actualPosition"], label="Elevation position", c=lineColors[colorCounter % nColors]
386 )
387 colorCounter += 1
388 ax1_twin.yaxis.set_major_formatter(FuncFormatter(tickFormatter))
389 ax1_twin.set_ylabel("Elevation (degrees)")
390 ax1.set_xticks([]) # remove x tick labels on the hidden upper x-axis
392 ax2_twin = ax2.twinx()
393 ax2.plot(azimuthData["actualTorque"], label="Azimuth torque", c=lineColors[colorCounter % nColors])
394 colorCounter += 1
395 ax2_twin.plot(
396 elevationData["actualTorque"], label="Elevation torque", c=lineColors[colorCounter % nColors]
397 )
398 colorCounter += 1
399 ax2.set_ylabel("Azimuth torque (Nm)")
400 ax2_twin.set_ylabel("Elevation torque (Nm)")
401 ax2.set_xlabel("Time (UTC)") # yes, it really is UTC, matplotlib converts this automatically!
403 # put the ticks at an angle, and right align with the tick marks
404 ax2.set_xticks(ax2.get_xticks()) # needed to supress a user warning
405 xlabels = ax2.get_xticks()
406 ax2.set_xticklabels(xlabels, rotation=40, ha="right")
407 ax2.xaxis.set_major_locator(mdates.AutoDateLocator())
408 ax2.xaxis.set_major_formatter(mdates.DateFormatter("%H:%M:%S"))
410 if event.type.name == "TRACKING":
411 # returns a copy
412 clippedAzimuthData = clipDataToEvent(azimuthData, event, postPadding=TRACKING_RESIDUAL_TAIL_CLIP)
413 clippedElevationData = clipDataToEvent(elevationData, event, postPadding=TRACKING_RESIDUAL_TAIL_CLIP)
415 azError = clippedAzimuthData["azError"].values
416 elError = clippedElevationData["elError"].values
417 elVals = clippedElevationData["actualPosition"].values
418 if doFilterResiduals:
419 # Filtering out bad values
420 nReplacedAz = filterBadValues(azError, maxDelta)
421 nReplacedEl = filterBadValues(elError, maxDelta)
422 clippedAzimuthData["azError"] = azError
423 clippedElevationData["elError"] = elError
424 # Calculate RMS
425 az_rms = np.sqrt(np.mean(azError * azError))
426 el_rms = np.sqrt(np.mean(elError * elError))
428 # Calculate Image impact RMS
429 # We are less sensitive to Az errors near the zenith
430 image_az_rms = az_rms * np.cos(elVals[0] * np.pi / 180.0)
431 image_el_rms = el_rms
432 image_impact_rms = np.sqrt(image_az_rms**2 + image_el_rms**2)
433 ax1p5.plot(
434 clippedAzimuthData["azError"],
435 label="Azimuth tracking error",
436 c=lineColors[colorCounter % nColors],
437 )
438 colorCounter += 1
439 ax1p5.plot(
440 clippedElevationData["elError"],
441 label="Elevation tracking error",
442 c=lineColors[colorCounter % nColors],
443 )
444 colorCounter += 1
445 ax1p5.axhline(0.01, ls="-.", color="black")
446 ax1p5.axhline(-0.01, ls="-.", color="black")
447 ax1p5.yaxis.set_major_formatter(FuncFormatter(tickFormatter))
448 ax1p5.set_ylabel("Tracking error (arcsec)")
449 ax1p5.set_xticks([]) # remove x tick labels on the hidden upper x-axis
450 ax1p5.set_ylim(-0.05, 0.05)
451 ax1p5.set_yticks([-0.04, -0.02, 0.0, 0.02, 0.04])
452 ax1p5.legend()
453 ax1p5.text(0.1, 0.9, f"Image impact RMS = {image_impact_rms:.3f} arcsec", transform=ax1p5.transAxes)
454 if doFilterResiduals:
455 ax1p5.text(
456 0.1,
457 0.8,
458 f"{nReplacedAz} bad azimuth values and {nReplacedEl} bad elevation values were replaced",
459 transform=ax1p5.transAxes,
460 )
462 if prePadding or postPadding:
463 # note the conversion to utc because the x-axis from the dataframe
464 # already got automagically converted when plotting before, so this is
465 # necessary for things to line up
466 ax1_twin.axvline(event.begin.utc.datetime, c="k", ls="--", alpha=0.5, label="Event begin/end")
467 ax1_twin.axvline(event.end.utc.datetime, c="k", ls="--", alpha=0.5)
468 # extend lines down across lower plot, but do not re-add label
469 ax2_twin.axvline(event.begin.utc.datetime, c="k", ls="--", alpha=0.5)
470 ax2_twin.axvline(event.end.utc.datetime, c="k", ls="--", alpha=0.5)
471 if ax1p5:
472 ax1p5.axvline(event.begin.utc.datetime, c="k", ls="--", alpha=0.5)
473 ax1p5.axvline(event.end.utc.datetime, c="k", ls="--", alpha=0.5)
475 for commandTime, command in commands.items():
476 plotTime = getPlotTime(commandTime)
477 ax1_twin.axvline(
478 plotTime, c=lineColors[colorCounter % nColors], ls="--", alpha=0.75, label=f"{command}"
479 )
480 # extend lines down across lower plot, but do not re-add label
481 ax2_twin.axvline(plotTime, c=lineColors[colorCounter % nColors], ls="--", alpha=0.75)
482 if ax1p5:
483 ax1p5.axvline(plotTime, c=lineColors[colorCounter % nColors], ls="--", alpha=0.75)
484 colorCounter += 1
486 # combine the legends and put inside the plot
487 handles1a, labels1a = ax1.get_legend_handles_labels()
488 handles1b, labels1b = ax1_twin.get_legend_handles_labels()
489 handles2a, labels2a = ax2.get_legend_handles_labels()
490 handles2b, labels2b = ax2_twin.get_legend_handles_labels()
492 handles = handles1a + handles1b + handles2a + handles2b
493 labels = labels1a + labels1b + labels2a + labels2b
494 # ax2 is "in front" of ax1 because it has the vlines plotted on it, and
495 # vlines are on ax2 so that they appear at the bottom of the legend, so
496 # make sure to plot the legend on ax2, otherwise the vlines will go on top
497 # of the otherwise-opaque legend.
498 ax1_twin.legend(handles, labels, facecolor="white", framealpha=1)
500 # Add title with the event name, type etc
501 dayObsStr = dayObsIntToString(event.dayObs)
502 title = (
503 # top line is the event title, the details go on the line below
504 f"{dayObsStr} - seqNum {event.seqNum} (version {event.version})"
505 f"\nDuration = {event.duration:.2f}s"
506 f" Event type: {event.type.name}"
507 f" End reason: {event.endReason.name}"
508 )
509 ax1_twin.set_title(title)
510 return fig
513def getCommandsDuringEvent(
514 client,
515 event,
516 commands=("raDecTarget"),
517 prePadding=0,
518 postPadding=0,
519 timeFormat="python",
520 log=None,
521 doLog=True,
522):
523 """Get the commands issued during an event.
525 Get the times at which the specified commands were issued during the event.
527 Parameters
528 ----------
529 client : `lsst_efd_client.efd_helper.EfdClient`
530 The EFD client to use.
531 event : `lsst.summit.utils.tmaUtils.TMAEvent`
532 The event to plot.
533 commands : `list` of `str`, optional
534 The commands or command aliases to look for. Defaults to
535 ['raDecTarget'].
536 prePadding : `float`, optional
537 The amount of time to pad the event with before the start time, in
538 seconds.
539 postPadding : `float`, optional
540 The amount of time to pad the event with after the end time, in
541 seconds.
542 timeFormat : `str`, optional
543 One of 'pandas' or 'astropy' or 'python'. If 'pandas', the dictionary
544 keys will be pandas timestamps, if 'astropy' they will be astropy times
545 and if 'python' they will be python datetimes.
546 log : `logging.Logger`, optional
547 The logger to use. If not specified, a new logger will be created if
548 needed.
549 doLog : `bool`, optional
550 Whether to log messages. Defaults to True.
552 Returns
553 -------
554 commandTimes : `dict` [`time`, `str`]
555 A dictionary of the times at which the commands where issued. The type
556 that `time` takes is determined by the format key, and defaults to
557 python datetime.
558 """
559 commands = list(ensure_iterable(commands))
560 fullCommands = [c if c not in COMMAND_ALIASES else COMMAND_ALIASES[c] for c in commands]
561 del commands # make sure we always use their full names
563 commandTimes = getCommands(
564 client,
565 fullCommands,
566 begin=event.begin,
567 end=event.end,
568 prePadding=prePadding,
569 postPadding=postPadding,
570 timeFormat=timeFormat,
571 )
573 if not commandTimes and doLog:
574 log = logging.getLogger(__name__)
575 log.info(f"Found no commands in {fullCommands} issued during event {event.seqNum}")
577 return commandTimes
580def _initializeTma(tma):
581 """Helper function to turn a TMA into a valid state for testing.
583 Do not call directly in normal usage or code, as this just arbitrarily
584 sets values to make the TMA valid.
586 Parameters
587 ----------
588 tma : `lsst.summit.utils.tmaUtils.TMAStateMachine`
589 The TMA state machine model to initialize.
590 """
591 tma._parts["azimuthInPosition"] = False
592 tma._parts["azimuthMotionState"] = AxisMotionState.STOPPED
593 tma._parts["azimuthSystemState"] = PowerState.ON
594 tma._parts["elevationInPosition"] = False
595 tma._parts["elevationMotionState"] = AxisMotionState.STOPPED
596 tma._parts["elevationSystemState"] = PowerState.ON
599@dataclass(kw_only=True, frozen=True)
600class TMAEvent:
601 """A movement event for the TMA.
603 Contains the dayObs on which the event occured, using the standard
604 observatory definition of the dayObs, and the sequence number of the event,
605 which is unique for each event on a given dayObs.
607 The event type can be either 'SLEWING' or 'TRACKING', defined as:
608 - SLEWING: some part of the TMA is in motion
609 - TRACKING: both axes are in position and tracking the sky
611 The end reason can be 'STOPPED', 'TRACKING', 'FAULT', 'SLEWING', or 'OFF'.
612 - SLEWING: The previous event was a TRACKING event, and one or more of
613 the TMA components either stopped being in position, or stopped
614 moving, or went into fault, or was turned off, and hence we are now
615 only slewing and no longer tracking the sky.
616 - TRACKING: the TMA started tracking the sky when it wasn't previously.
617 Usualy this would always be preceded by directly by a SLEWING
618 event, but this is not strictly true, as the EUI seems to be able
619 to make the TMA start tracking the sky without slewing first.
620 - STOPPED: the components of the TMA transitioned to the STOPPED state.
621 - FAULT: the TMA went into fault.
622 - OFF: the TMA components were turned off.
624 Note that this class is not intended to be instantiated directly, but
625 rather to be returned by the ``TMAEventMaker.getEvents()`` function.
627 Parameters
628 ----------
629 dayObs : `int`
630 The dayObs on which the event occured.
631 seqNum : `int`
632 The sequence number of the event,
633 type : `lsst.summit.utils.tmaUtils.TMAState`
634 The type of the event, either 'SLEWING' or 'TRACKING'.
635 endReason : `lsst.summit.utils.tmaUtils.TMAState`
636 The reason the event ended, either 'STOPPED', 'TRACKING', 'FAULT',
637 'SLEWING', or 'OFF'.
638 duration : `float`
639 The duration of the event, in seconds.
640 begin : `astropy.time.Time`
641 The time the event began.
642 end : `astropy.time.Time`
643 The time the event ended.
644 blockInfos : `list` of `lsst.summit.utils.tmaUtils.BlockInfo`, or `None`
645 The block infomation, if any, relating to the event. Could be `None`,
646 or one or more block informations.
647 version : `int`
648 The version of the TMAEvent class. Equality between events is only
649 valid for a given version of the class. If the class definition
650 changes, the time ranges can change, and hence the equality between
651 events is ``False``.
652 _startRow : `int`
653 The first row in the merged EFD data which is part of the event.
654 _endRow : `int`
655 The last row in the merged EFD data which is part of the event.
656 """
658 dayObs: int
659 seqNum: int
660 type: str # can be 'SLEWING', 'TRACKING'
661 endReason: str # can be 'STOPPED', 'TRACKING', 'FAULT', 'SLEWING', 'OFF'
662 duration: float # seconds
663 begin: Time
664 end: Time
665 blockInfos: list = field(default_factory=list)
666 version: int = 0 # update this number any time a code change which could change event definitions is made
667 _startRow: int
668 _endRow: int
670 def __lt__(self, other):
671 if self.version != other.version:
672 raise ValueError(
673 f"Cannot compare TMAEvents with different versions: {self.version} != {other.version}"
674 )
675 if self.dayObs < other.dayObs:
676 return True
677 elif self.dayObs == other.dayObs:
678 return self.seqNum < other.seqNum
679 return False
681 def __repr__(self):
682 return (
683 f"TMAEvent(dayObs={self.dayObs}, seqNum={self.seqNum}, type={self.type!r},"
684 f" endReason={self.endReason!r}, duration={self.duration}, begin={self.begin!r},"
685 f" end={self.end!r}"
686 )
688 def __hash__(self):
689 # deliberately don't hash the blockInfos here, as they are not
690 # a core part of the event itself, and are listy and cause problems
691 return hash(
692 (
693 self.dayObs,
694 self.seqNum,
695 self.type,
696 self.endReason,
697 self.duration,
698 self.begin,
699 self.end,
700 self.version,
701 self._startRow,
702 self._endRow,
703 )
704 )
706 def _ipython_display_(self):
707 print(self.__str__())
709 def __str__(self):
710 def indent(string):
711 return "\n" + "\n".join([" " + s for s in string.splitlines()])
713 blockInfoStr = "None"
714 if self.blockInfos is not None:
715 blockInfoStr = "".join(indent(str(i)) for i in self.blockInfos)
717 return (
718 f"dayObs: {self.dayObs}\n"
719 f"seqNum: {self.seqNum}\n"
720 f"type: {self.type.name}\n"
721 f"endReason: {self.endReason.name}\n"
722 f"duration: {self.duration}\n"
723 f"begin: {self.begin!r}\n"
724 f"end: {self.end!r}\n"
725 f"blockInfos: {blockInfoStr}"
726 )
728 def associatedWith(self, block=None, blockSeqNum=None, ticket=None, salIndex=None):
729 """Check whether an event is associated with a set of parameters.
731 Check if an event is associated with a specific block and/or ticket
732 and/or salIndex. All specified parameters must match for the function
733 to return True. If checking if an event is in a block, the blockSeqNum
734 can also be specified to identify events which related to a given
735 running the specified block.
737 Parameters
738 ----------
739 block : `int`, optional
740 The block number to check for.
741 blockSeqNum : `int`, optional
742 The block sequence number to check for, if the block is specified.
743 ticket : `str`, optional
744 The ticket number to check for.
745 salIndex : `int`, optional
746 The salIndex to check for.
748 Returns
749 -------
750 relates : `bool`
751 Whether the event is associated with the specified block, ticket,
752 and salIndex.
753 """
754 if all([block is None, ticket is None, salIndex is None]):
755 raise ValueError("Must specify at least one of block, ticket, or salIndex")
757 if blockSeqNum is not None and block is None:
758 raise ValueError("block must be specified if blockSeqNum is specified")
760 for blockInfo in self.blockInfos:
761 # "X is None or" is used for each parameter to allow it to be None
762 # in the kwargs
763 blockMatches = False
764 if block is not None:
765 if blockSeqNum is None and blockInfo.blockNumber == block:
766 blockMatches = True
767 elif (
768 blockSeqNum is not None
769 and blockInfo.blockNumber == block
770 and blockInfo.seqNum == blockSeqNum
771 ):
772 blockMatches = True
773 else:
774 blockMatches = True # no block specified at all, so it matches
776 salIndexMatches = salIndex is None or salIndex in blockInfo.salIndices
777 ticketMatches = ticket is None or ticket in blockInfo.tickets
779 if blockMatches and salIndexMatches and ticketMatches:
780 return True
782 return False
785class TMAState(enum.IntEnum):
786 """Overall state of the TMA.
788 States are defined as follows:
790 UNINITIALIZED
791 We have not yet got data for all relevant components, so the overall
792 state is undefined.
793 STOPPED
794 All components are on, and none are moving.
795 TRACKING
796 We are tracking the sky.
797 SLEWING
798 One or more components are moving, and one or more are not tracking the
799 sky. This should probably be called MOVING, as it includes: slewing,
800 MOVING_POINT_TO_POINT, and JOGGING.
801 FAULT
802 All (if engineeringMode) or any (if not engineeringMode) components are
803 in fault.
804 OFF
805 All components are off.
806 """
808 UNINITIALIZED = -1
809 STOPPED = 0
810 TRACKING = 1
811 SLEWING = 2
812 FAULT = 3
813 OFF = 4
815 def __repr__(self):
816 return f"TMAState.{self.name}"
819def getAxisAndType(rowFor):
820 """Get the axis the data relates to, and the type of data it contains.
822 Parameters
823 ----------
824 rowFor : `str`
825 The column in the dataframe denoting what this row is for, e.g.
826 "elevationMotionState" or "azimuthInPosition", etc.
828 Returns
829 -------
830 axis : `str`
831 The axis the row is for, e.g. "azimuth", "elevation".
832 rowType : `str`
833 The type of the row, e.g. "MotionState", "SystemState", "InPosition".
834 """
835 regex = r"(azimuth|elevation)(InPosition|MotionState|SystemState)$" # matches the end of the line
836 matches = re.search(regex, rowFor)
837 if matches is None:
838 raise ValueError(f"Could not parse axis and rowType from {rowFor=}")
839 axis = matches.group(1)
840 rowType = matches.group(2)
842 assert rowFor.endswith(f"{axis}{rowType}")
843 return axis, rowType
846class ListViewOfDict:
847 """A class to allow making lists which contain references to an underlying
848 dictionary.
850 Normally, making a list of items from a dictionary would make a copy of the
851 items, but this class allows making a list which contains references to the
852 underlying dictionary items themselves. This is useful for making a list of
853 components, such that they can be manipulated in their logical sets.
854 """
856 def __init__(self, underlyingDictionary, keysToLink):
857 self.dictionary = underlyingDictionary
858 self.keys = keysToLink
860 def __getitem__(self, index):
861 return self.dictionary[self.keys[index]]
863 def __setitem__(self, index, value):
864 self.dictionary[self.keys[index]] = value
866 def __len__(self):
867 return len(self.keys)
870class TMAStateMachine:
871 """A state machine model of the TMA.
873 Note that this is currently only implemented for the azimuth and elevation
874 axes, but will be extended to include the rotator in the future.
876 Note that when used for event generation, changing ``engineeringMode`` to
877 False might change the resulting list of events, and that if the TMA moves
878 with some axis in fault, then these events will be missed. It is therefore
879 thought that ``engineeringMode=True`` should always be used when generating
880 events. The option, however, is there for completeness, as this will be
881 useful for knowing is the CSC would consider the TMA to be in fault in the
882 general case.
884 Parameters
885 ----------
886 engineeringMode : `bool`, optional
887 Whether the TMA is in engineering mode. Defaults to True. If False,
888 then the TMA will be in fault if any component is in fault. If True,
889 then the TMA will be in fault only if all components are in fault.
890 debug : `bool`, optional
891 Whether to log debug messages. Defaults to False.
892 """
894 _UNINITIALIZED_VALUE: int = -999
896 def __init__(self, engineeringMode=True, debug=False):
897 self.engineeringMode = engineeringMode
898 self.log = logging.getLogger("lsst.summit.utils.tmaUtils.TMA")
899 if debug:
900 self.log.level = logging.DEBUG
901 self._mostRecentRowTime = -1
903 # the actual components of the TMA
904 self._parts = {
905 "azimuthInPosition": self._UNINITIALIZED_VALUE,
906 "azimuthMotionState": self._UNINITIALIZED_VALUE,
907 "azimuthSystemState": self._UNINITIALIZED_VALUE,
908 "elevationInPosition": self._UNINITIALIZED_VALUE,
909 "elevationMotionState": self._UNINITIALIZED_VALUE,
910 "elevationSystemState": self._UNINITIALIZED_VALUE,
911 }
912 systemKeys = ["azimuthSystemState", "elevationSystemState"]
913 positionKeys = ["azimuthInPosition", "elevationInPosition"]
914 motionKeys = ["azimuthMotionState", "elevationMotionState"]
916 # references to the _parts as conceptual groupings
917 self.system = ListViewOfDict(self._parts, systemKeys)
918 self.motion = ListViewOfDict(self._parts, motionKeys)
919 self.inPosition = ListViewOfDict(self._parts, positionKeys)
921 # tuples of states for state collapsing. Note that STOP_LIKE +
922 # MOVING_LIKE must cover the full set of AxisMotionState enums
923 self.STOP_LIKE = (AxisMotionState.STOPPING, AxisMotionState.STOPPED, AxisMotionState.TRACKING_PAUSED)
924 self.MOVING_LIKE = (
925 AxisMotionState.MOVING_POINT_TO_POINT,
926 AxisMotionState.JOGGING,
927 AxisMotionState.TRACKING,
928 )
929 # Likewise, ON_LIKE + OFF_LIKE must cover the full set of PowerState
930 # enums
931 self.OFF_LIKE = (PowerState.OFF, PowerState.TURNING_OFF)
932 self.ON_LIKE = (PowerState.ON, PowerState.TURNING_ON)
933 self.FAULT_LIKE = (PowerState.FAULT,) # note the trailing comma - this must be an iterable
935 def apply(self, row):
936 """Apply a row of data to the TMA state.
938 Checks that the row contains data for a later time than any data
939 previously applied, and applies the relevant column entry to the
940 relevant component.
942 Parameters
943 ----------
944 row : `pd.Series`
945 The row of data to apply to the state machine.
946 """
947 timestamp = row["private_efdStamp"]
948 if timestamp < self._mostRecentRowTime: # NB equals is OK, technically, though it never happens
949 raise ValueError(
950 "TMA evolution must be monotonic increasing in time, tried to apply a row which"
951 " predates the most previous one"
952 )
953 self._mostRecentRowTime = timestamp
955 rowFor = row["rowFor"] # e.g. elevationMotionState
956 axis, rowType = getAxisAndType(rowFor) # e.g. elevation, MotionState
957 value = self._getRowPayload(row, rowType, rowFor)
958 self.log.debug(f"Setting {rowFor} to {repr(value)}")
959 self._parts[rowFor] = value
960 try:
961 # touch the state property as this executes the sieving, to make
962 # sure we don't fall through the sieve at any point in time
963 _ = self.state
964 except RuntimeError as e:
965 # improve error reporting, but always reraise this, as this is a
966 # full-blown failure
967 raise RuntimeError(f"Failed to apply {value} to {axis}{rowType} with state {self._parts}") from e
969 def _getRowPayload(self, row, rowType, rowFor):
970 """Get the relevant value from the row.
972 Given the row, and which component it relates to, get the relevant
973 value, as a bool or cast to the appropriate enum class.
975 Parameters
976 ----------
977 row : `pd.Series`
978 The row of data from the dataframe.
979 rowType : `str`
980 The type of the row, e.g. "MotionState", "SystemState",
981 "InPosition".
982 rowFor : `str`
983 The component the row is for, e.g. "azimuth", "elevation".
985 Returns
986 -------
987 value : `bool` or `enum`
988 The value of the row, as a bool or enum, depending on the
989 component, cast to the appropriate enum class or bool.
990 """
991 match rowType:
992 case "MotionState":
993 value = row[f"state_{rowFor}"]
994 return AxisMotionState(value)
995 case "SystemState":
996 value = row[f"powerState_{rowFor}"]
997 return PowerState(value)
998 case "InPosition":
999 value = row[f"inPosition_{rowFor}"]
1000 return bool(value)
1001 case _:
1002 raise ValueError(f"Failed to get row payload with {rowType=} and {row=}")
1004 @property
1005 def _isValid(self):
1006 """Has the TMA had a value applied to all its components?
1008 If any component has not yet had a value applied, the TMA is not valid,
1009 as those components will be in an unknown state.
1011 Returns
1012 -------
1013 isValid : `bool`
1014 Whether the TMA is fully initialized.
1015 """
1016 return not any([v == self._UNINITIALIZED_VALUE for v in self._parts.values()])
1018 # state inspection properties - a high level way of inspecting the state as
1019 # an API
1020 @property
1021 def isMoving(self):
1022 return self.state in [TMAState.TRACKING, TMAState.SLEWING]
1024 @property
1025 def isNotMoving(self):
1026 return not self.isMoving
1028 @property
1029 def isTracking(self):
1030 return self.state == TMAState.TRACKING
1032 @property
1033 def isSlewing(self):
1034 return self.state == TMAState.SLEWING
1036 @property
1037 def canMove(self):
1038 badStates = [PowerState.OFF, PowerState.TURNING_OFF, PowerState.FAULT, PowerState.UNKNOWN]
1039 return bool(
1040 self._isValid
1041 and self._parts["azimuthSystemState"] not in badStates
1042 and self._parts["elevationSystemState"] not in badStates
1043 )
1045 # Axis inspection properties, designed for internal use. These return
1046 # iterables so that they can be used in any() and all() calls, which make
1047 # the logic much easier to read, e.g. to see if anything is moving, we can
1048 # write `if not any(_axisInMotion):`
1049 @property
1050 def _axesInFault(self):
1051 return [x in self.FAULT_LIKE for x in self.system]
1053 @property
1054 def _axesOff(self):
1055 return [x in self.OFF_LIKE for x in self.system]
1057 @property
1058 def _axesOn(self):
1059 return [not x for x in self._axesOn]
1061 @property
1062 def _axesInMotion(self):
1063 return [x in self.MOVING_LIKE for x in self.motion]
1065 @property
1066 def _axesTRACKING(self):
1067 """Note this is deliberately named _axesTRACKING and not _axesTracking
1068 to make it clear that this is the AxisMotionState type of TRACKING and
1069 not the normal conceptual notion of tracking (the sky, i.e. as opposed
1070 to slewing).
1071 """
1072 return [x == AxisMotionState.TRACKING for x in self.motion]
1074 @property
1075 def _axesInPosition(self):
1076 return [x is True for x in self.inPosition]
1078 @property
1079 def state(self):
1080 """The overall state of the TMA.
1082 Note that this is both a property, and also the method which applies
1083 the logic sieve to determine the state at a given point in time.
1085 Returns
1086 -------
1087 state : `lsst.summit.utils.tmaUtils.TMAState`
1088 The overall state of the TMA.
1089 """
1090 # first, check we're valid, and if not, return UNINITIALIZED state, as
1091 # things are unknown
1092 if not self._isValid:
1093 return TMAState.UNINITIALIZED
1095 # if we're not in engineering mode, i.e. we're under normal CSC
1096 # control, then if anything is in fault, we're in fault. If we're
1097 # engineering then some axes will move when others are in fault
1098 if not self.engineeringMode:
1099 if any(self._axesInFault):
1100 return TMAState.FAULT
1101 else:
1102 # we're in engineering mode, so return fault state if ALL are in
1103 # fault
1104 if all(self._axesInFault):
1105 return TMAState.FAULT
1107 # if all axes are off, the TMA is OFF
1108 if all(self._axesOff):
1109 return TMAState.OFF
1111 # we know we're valid and at least some axes are not off, so see if
1112 # we're in motion if no axes are moving, we're stopped
1113 if not any(self._axesInMotion):
1114 return TMAState.STOPPED
1116 # now we know we're initialized, and that at least one axis is moving
1117 # so check axes for motion and in position. If all axes are tracking
1118 # and all are in position, we're tracking the sky
1119 if all(self._axesTRACKING) and all(self._axesInPosition):
1120 return TMAState.TRACKING
1122 # we now know explicitly that not everything is in position, so we no
1123 # longer need to check that. We do actually know that something is in
1124 # motion, but confirm that's the case and return SLEWING
1125 if any(self._axesInMotion):
1126 return TMAState.SLEWING
1128 # if we want to differentiate between MOVING_POINT_TO_POINT moves,
1129 # JOGGING moves and regular slews, the logic in the step above needs to
1130 # be changed and the new steps added here.
1132 raise RuntimeError("State error: fell through the state sieve - rewrite your logic!")
1135class TMAEventMaker:
1136 """A class to create per-dayObs TMAEvents for the TMA's movements.
1138 If this class is being used in tests, make sure to pass the EFD client in,
1139 and create it with `makeEfdClient(testing=True)`. This ensures that the
1140 USDF EFD is "used" as this is the EFD which has the recorded data available
1141 in the test suite via `vcr`.
1143 Example usage:
1144 >>> dayObs = 20230630
1145 >>> eventMaker = TMAEventMaker()
1146 >>> events = eventMaker.getEvents(dayObs)
1147 >>> print(f'Found {len(events)} for {dayObs=}')
1149 Parameters
1150 ----------
1151 client : `lsst_efd_client.efd_helper.EfdClient`, optional
1152 The EFD client to use, created if not provided.
1153 """
1155 # the topics which need logical combination to determine the overall mount
1156 # state. Will need updating as new components are added to the system.
1158 # relevant column: 'state'
1159 _movingComponents = [
1160 "lsst.sal.MTMount.logevent_azimuthMotionState",
1161 "lsst.sal.MTMount.logevent_elevationMotionState",
1162 ]
1164 # relevant column: 'inPosition'
1165 _inPositionComponents = [
1166 "lsst.sal.MTMount.logevent_azimuthInPosition",
1167 "lsst.sal.MTMount.logevent_elevationInPosition",
1168 ]
1170 # the components which, if in fault, put the TMA into fault
1171 # relevant column: 'powerState'
1172 _stateComponents = [
1173 "lsst.sal.MTMount.logevent_azimuthSystemState",
1174 "lsst.sal.MTMount.logevent_elevationSystemState",
1175 ]
1177 def __init__(self, client=None):
1178 if client is not None:
1179 self.client = client
1180 else:
1181 self.client = makeEfdClient()
1182 self.log = logging.getLogger(__name__)
1183 self._data = {}
1185 @dataclass(frozen=True)
1186 class ParsedState:
1187 eventStart: Time
1188 eventEnd: int
1189 previousState: TMAState
1190 state: TMAState
1192 @staticmethod
1193 def isToday(dayObs):
1194 """Find out if the specified dayObs is today, or in the past.
1196 If the day is today, the function returns ``True``, if it is in the
1197 past it returns ``False``. If the day is in the future, a
1198 ``ValueError`` is raised, as this indicates there is likely an
1199 off-by-one type error somewhere in the logic.
1201 Parameters
1202 ----------
1203 dayObs : `int`
1204 The dayObs to check, in the format YYYYMMDD.
1206 Returns
1207 -------
1208 isToday : `bool`
1209 ``True`` if the dayObs is today, ``False`` if it is in the past.
1211 Raises
1212 ValueError: if the dayObs is in the future.
1213 """
1214 todayDayObs = getCurrentDayObs_int()
1215 if dayObs == todayDayObs:
1216 return True
1217 if dayObs > todayDayObs:
1218 raise ValueError("dayObs is in the future")
1219 return False
1221 @staticmethod
1222 def _shortName(topic):
1223 """Get the short name of a topic.
1225 Parameters
1226 ----------
1227 topic : `str`
1228 The topic to get the short name of.
1230 Returns
1231 -------
1232 shortName : `str`
1233 The short name of the topic, e.g. 'azimuthInPosition'
1234 """
1235 # get, for example 'azimuthInPosition' from
1236 # lsst.sal.MTMount.logevent_azimuthInPosition
1237 return topic.split("_")[-1]
1239 def _mergeData(self, data):
1240 """Merge a dict of dataframes based on private_efdStamp, recording
1241 where each row came from.
1243 Given a dict or dataframes, keyed by topic, merge them into a single
1244 dataframe, adding a column to record which topic each row came from.
1246 Parameters
1247 ----------
1248 data : `dict` of `str` : `pd.DataFrame`
1249 The dataframes to merge.
1251 Returns
1252 -------
1253 merged : `pd.DataFrame`
1254 The merged dataframe.
1255 """
1256 excludeColumns = ["private_efdStamp", "rowFor"]
1258 mergeArgs = {
1259 "how": "outer",
1260 "sort": True,
1261 }
1263 merged = None
1264 originalRowCounter = 0
1266 # Iterate over the keys and merge the corresponding DataFrames
1267 for key, df in data.items():
1268 if df.empty:
1269 # Must skip the df if it's empty, otherwise the merge will fail
1270 # due to lack of private_efdStamp. Because other axes might
1271 # still be in motion, so we still want to merge what we have
1272 continue
1274 originalRowCounter += len(df)
1275 component = self._shortName(key) # Add suffix to column names to identify the source
1276 suffix = "_" + component
1278 df["rowFor"] = component
1280 columnsToSuffix = [col for col in df.columns if col not in excludeColumns]
1281 df_to_suffix = df[columnsToSuffix].add_suffix(suffix)
1282 df = pd.concat([df[excludeColumns], df_to_suffix], axis=1)
1284 if merged is None:
1285 merged = df.copy()
1286 else:
1287 merged = pd.merge(merged, df, **mergeArgs)
1289 merged = merged.loc[:, ~merged.columns.duplicated()] # Remove duplicate columns after merge
1291 if len(merged) != originalRowCounter:
1292 self.log.warning(
1293 "Merged data has a different number of rows to the original data, some"
1294 " timestamps (rows) will contain more than one piece of actual information."
1295 )
1297 # if the index is still a DatetimeIndex here then we didn't actually
1298 # merge any data, so there is only data from a single component.
1299 # This is likely to result in no events, but not necessarily, and for
1300 # generality, instead we convert to a range index to ensure consistency
1301 # in the returned data, and allow processing to continue.
1302 if isinstance(merged.index, pd.DatetimeIndex):
1303 self.log.warning("Data was only found for a single component in the EFD.")
1304 merged.reset_index(drop=True, inplace=True)
1306 return merged
1308 def getEvent(self, dayObs, seqNum):
1309 """Get a specific event for a given dayObs and seqNum.
1311 Repeated calls for the same ``dayObs`` will use the cached data if the
1312 day is in the past, and so will be much quicker. If the ``dayObs`` is
1313 the current day then the EFD will be queried for new data for each
1314 call, so a call which returns ``None`` on the first try might return an
1315 event on the next, if the TMA is still moving and thus generating
1316 events.
1318 Parameters
1319 ----------
1320 dayObs : `int`
1321 The dayObs to get the event for.
1322 seqNum : `int`
1323 The sequence number of the event to get.
1325 Returns
1326 -------
1327 event : `lsst.summit.utils.tmaUtils.TMAEvent`
1328 The event for the specified dayObs and seqNum, or `None` if the
1329 event was not found.
1330 """
1331 events = self.getEvents(dayObs)
1332 if seqNum <= len(events):
1333 event = events[seqNum]
1334 if event.seqNum != seqNum:
1335 # it's zero-indexed and contiguous so this must be true but
1336 # a sanity check doesn't hurt.
1337 raise AssertionError(f"Event sequence number mismatch: {event.seqNum} != {seqNum}")
1338 return event
1339 else:
1340 self.log.warning(f"Event {seqNum} not found for {dayObs}")
1341 return None
1343 def getEvents(self, dayObs, addBlockInfo=True):
1344 """Get the TMA events for the specified dayObs.
1346 Gets the required mount data from the cache or the EFD as required,
1347 handling whether we're working with live vs historical data. The
1348 dataframes from the EFD is merged and applied to the TMAStateMachine,
1349 and that series of state changes is used to generate a list of
1350 TmaEvents for the day's data.
1352 If the data is for the current day, i.e. if new events can potentially
1353 land, then if the last event is "open" (meaning that the TMA appears to
1354 be in motion and thus the event is growing with time), then that event
1355 is excluded from the event list as it is expected to be changing with
1356 time, and will likely close eventually. However, if that situation
1357 occurs on a day in the past, then that event can never close, and the
1358 event is therefore included, but a warning about the open event is
1359 logged.
1361 Parameters
1362 ----------
1363 dayObs : `int`
1364 The dayObs for which to get the events.
1365 addBlockInfo : `bool`, optional
1366 Whether to add block information to the events. This allows
1367 skipping this step for speed when generating events for purposes
1368 which don't need block information.
1370 Returns
1371 -------
1372 events : `list` of `lsst.summit.utils.tmaUtils.TMAState`
1373 The events for the specified dayObs.
1374 """
1375 workingLive = self.isToday(dayObs)
1376 data = None
1378 if workingLive:
1379 # it's potentially updating data, so we must update the date
1380 # regarless of whether we have it already or not
1381 self.log.info(f"Updating mount data for {dayObs} from the EFD")
1382 self._getEfdDataForDayObs(dayObs)
1383 data = self._data[dayObs]
1384 elif dayObs in self._data:
1385 # data is in the cache and it's not being updated, so use it
1386 data = self._data[dayObs]
1387 elif dayObs not in self._data:
1388 # we don't have the data yet, but it's not growing, so put it in
1389 # the cache and use it from there
1390 self.log.info(f"Retrieving mount data for {dayObs} from the EFD")
1391 self._getEfdDataForDayObs(dayObs)
1392 data = self._data[dayObs]
1393 else:
1394 raise RuntimeError("This should never happen")
1396 # if we don't have something to work with, log a warning and return
1397 if not self.dataFound(data):
1398 self.log.warning(f"No EFD data found for {dayObs=}")
1399 return []
1401 # applies the data to the state machine, and generates events from the
1402 # series of states which results
1403 events = self._calculateEventsFromMergedData(
1404 data, dayObs, dataIsForCurrentDay=workingLive, addBlockInfo=addBlockInfo
1405 )
1406 if not events:
1407 self.log.warning(f"Failed to calculate any events for {dayObs=} despite EFD data existing!")
1408 return events
1410 @staticmethod
1411 def dataFound(data):
1412 """Check if any data was found.
1414 Parameters
1415 ----------
1416 data : `pd.DataFrame`
1417 The merged dataframe to check.
1419 Returns
1420 -------
1421 dataFound : `bool`
1422 Whether data was found.
1423 """
1424 # You can't just compare to with data == NO_DATA_SENTINEL because
1425 # `data` is usually a dataframe, and you can't compare a dataframe to a
1426 # string directly.
1427 return not (isinstance(data, str) and data == NO_DATA_SENTINEL)
1429 def _getEfdDataForDayObs(self, dayObs):
1430 """Get the EFD data for the specified dayObs and store it in the cache.
1432 Gets the EFD data for all components, as a dict of dataframes keyed by
1433 component name. These are then merged into a single dataframe in time
1434 order, based on each row's `private_efdStamp`. This is then stored in
1435 self._data[dayObs].
1437 If no data is found, the value is set to ``NO_DATA_SENTINEL`` to
1438 differentiate this from ``None``, as this is what you'd get if you
1439 queried the cache with `self._data.get(dayObs)`. It also marks that we
1440 have already queried this day.
1442 Parameters
1443 ----------
1444 dayObs : `int`
1445 The dayObs to query.
1446 """
1447 data = {}
1448 for component in itertools.chain(
1449 self._movingComponents, self._inPositionComponents, self._stateComponents
1450 ):
1451 data[component] = getEfdData(self.client, component, dayObs=dayObs, warn=False)
1452 self.log.debug(f"Found {len(data[component])} for {component}")
1454 if all(dataframe.empty for dataframe in data.values()):
1455 # if every single dataframe is empty, set the sentinel and don't
1456 # try to merge anything, otherwise merge all the data we found
1457 self.log.debug(f"No data found for {dayObs=}")
1458 # a sentinel value that's not None
1459 self._data[dayObs] = NO_DATA_SENTINEL
1460 else:
1461 merged = self._mergeData(data)
1462 self._data[dayObs] = merged
1464 def _calculateEventsFromMergedData(self, data, dayObs, dataIsForCurrentDay, addBlockInfo):
1465 """Calculate the list of events from the merged data.
1467 Runs the merged data, row by row, through the TMA state machine (with
1468 ``tma.apply``) to get the overall TMA state at each row, building a
1469 dict of these states, keyed by row number.
1471 This time-series of TMA states are then looped over (in
1472 `_statesToEventTuples`), building a list of tuples representing the
1473 start and end of each event, the type of the event, and the reason for
1474 the event ending.
1476 This list of tuples is then passed to ``_makeEventsFromStateTuples``,
1477 which actually creates the ``TMAEvent`` objects.
1479 Parameters
1480 ----------
1481 data : `pd.DataFrame`
1482 The merged dataframe to use.
1483 dayObs : `int`
1484 The dayObs for the data.
1485 dataIsForCurrentDay : `bool`
1486 Whether the data is for the current day. Determines whether to
1487 allow an open last event or not.
1488 addBlockInfo : `bool`
1489 Whether to add block information to the events. This allows
1490 skipping this step for speed when generating events for purposes
1491 which don't need block information.
1493 Returns
1494 -------
1495 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
1496 The events for the specified dayObs.
1497 """
1498 engineeringMode = True
1499 tma = TMAStateMachine(engineeringMode=engineeringMode)
1501 # For now, we assume that the TMA starts each day able to move, but
1502 # stationary. If this turns out to cause problems, we will need to
1503 # change to loading data from the previous day(s), and looking back
1504 # through it in time until a state change has been found for every
1505 # axis. For now though, Bruno et. al think this is acceptable and
1506 # preferable.
1507 _initializeTma(tma)
1509 tmaStates = {}
1510 for rowNum, row in data.iterrows():
1511 tma.apply(row)
1512 tmaStates[rowNum] = tma.state
1514 stateTuples = self._statesToEventTuples(tmaStates, dataIsForCurrentDay)
1515 events = self._makeEventsFromStateTuples(stateTuples, dayObs, data)
1516 if addBlockInfo:
1517 self.addBlockDataToEvents(dayObs, events)
1518 return events
1520 def _statesToEventTuples(self, states, dataIsForCurrentDay):
1521 """Get the event-tuples from the dictionary of TMAStates.
1523 Chunks the states into blocks of the same state, so that we can create
1524 an event for each block in `_makeEventsFromStateTuples`. Off-type
1525 states are skipped over, with each event starting when the telescope
1526 next resumes motion or changes to a different type of motion state,
1527 i.e. from non-tracking type movement (MOVE_POINT_TO_POINT, JOGGING,
1528 TRACKING-but-not-in-position, i.e. slewing) to a tracking type
1529 movement, or vice versa.
1531 Parameters
1532 ----------
1533 states : `dict` of `int` : `lsst.summit.utils.tmaUtils.TMAState`
1534 The states of the TMA, keyed by row number.
1535 dataIsForCurrentDay : `bool`
1536 Whether the data is for the current day. Determines whether to
1537 allow and open last event or not.
1539 Returns
1540 -------
1541 parsedStates : `list` of `tuple`
1542 The parsed states, as a list of tuples of the form:
1543 ``(eventStart, eventEnd, eventType, endReason)``
1544 """
1545 # Consider rewriting this with states as a list and using pop(0)?
1546 skipStates = (TMAState.STOPPED, TMAState.OFF, TMAState.FAULT)
1548 parsedStates = []
1549 eventStart = None
1550 rowNum = 0
1551 nRows = len(states)
1552 while rowNum < nRows:
1553 previousState = None
1554 state = states[rowNum]
1555 # if we're not in an event, fast forward through off-like rows
1556 # until a new event starts
1557 if eventStart is None and state in skipStates:
1558 rowNum += 1
1559 continue
1561 # we've started a new event, so walk through it and find the end
1562 eventStart = rowNum
1563 previousState = state
1564 rowNum += 1 # move to the next row before starting the while loop
1565 if rowNum == nRows:
1566 # we've reached the end of the data, and we're still in an
1567 # event, so don't return this presumably in-progress event
1568 self.log.warning("Reached the end of the data while starting a new event")
1569 break
1570 state = states[rowNum]
1571 while state == previousState:
1572 rowNum += 1
1573 if rowNum == nRows:
1574 break
1575 state = states[rowNum]
1576 parsedStates.append(
1577 self.ParsedState(
1578 eventStart=eventStart, eventEnd=rowNum, previousState=previousState, state=state
1579 )
1580 )
1581 if state in skipStates:
1582 eventStart = None
1584 # done parsing, just check the last event is valid
1585 if parsedStates: # ensure we have at least one event
1586 lastEvent = parsedStates[-1]
1587 if lastEvent.eventEnd == nRows:
1588 # Generally, you *want* the timespan for an event to be the
1589 # first row of the next event, because you were in that state
1590 # right up until that state change. However, if that event is
1591 # a) the last one of the day and b) runs right up until the end
1592 # of the dataframe, then there isn't another row, so this will
1593 # overrun the array.
1594 #
1595 # If the data is for the current day then this isn't a worry,
1596 # as we're likely still taking data, and this event will likely
1597 # close yet, so we don't issue a warning, and simply drop the
1598 # event from the list.
1600 # However, if the data is for a past day then no new data will
1601 # come to close the event, so allow the event to be "open", and
1602 # issue a warning
1603 if dataIsForCurrentDay:
1604 self.log.info("Discarding open (likely in-progess) final event from current day's events")
1605 parsedStates = parsedStates[:-1]
1606 else:
1607 self.log.warning("Last event ends open, forcing it to end at end of the day's data")
1608 # it's a tuple, so (deliberately) awkward to modify
1609 parsedStates[-1] = self.ParsedState(
1610 eventStart=lastEvent.eventStart,
1611 eventEnd=lastEvent.eventEnd - 1,
1612 previousState=lastEvent.previousState,
1613 state=lastEvent.state,
1614 )
1616 return parsedStates
1618 def addBlockDataToEvents(self, dayObs, events):
1619 """Find all the block data in the EFD for the specified events.
1621 Finds all the block data in the EFD relating to the events, parses it,
1622 from the rows of the dataframe, and adds it to the events in place.
1624 Parameters
1625 ----------
1626 events : `lsst.summit.utils.tmaUtils.TMAEvent` or
1627 `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
1628 One or more events to get the block data for.
1629 """
1630 try:
1631 blockParser = BlockParser(dayObs, client=self.client)
1632 except Exception as e:
1633 # adding the block data should never cause a failure so if we can't
1634 # get the block data, log a warning and return. It is, however,
1635 # never expected, so use log.exception to get the full traceback
1636 # and scare users so it gets reported
1637 self.log.exception(f"Failed to parse block data for {dayObs=}, {e}")
1638 return
1639 blocks = blockParser.getBlockNums()
1640 blockDict = {}
1641 for block in blocks:
1642 blockDict[block] = blockParser.getSeqNums(block)
1644 for block, seqNums in blockDict.items():
1645 for seqNum in seqNums:
1646 blockInfo = blockParser.getBlockInfo(block=block, seqNum=seqNum)
1648 relatedEvents = blockParser.getEventsForBlock(events, block=block, seqNum=seqNum)
1649 for event in relatedEvents:
1650 toSet = [blockInfo]
1651 if event.blockInfos is not None:
1652 existingInfo = event.blockInfos
1653 existingInfo.append(blockInfo)
1654 toSet = existingInfo
1656 # Add the blockInfo to the TMAEvent. Because this is a
1657 # frozen dataclass, use object.__setattr__ to set the
1658 # attribute. This is the correct way to set a frozen
1659 # dataclass attribute after creation.
1660 object.__setattr__(event, "blockInfos", toSet)
1662 def _makeEventsFromStateTuples(self, states, dayObs, data):
1663 """For the list of state-tuples, create a list of ``TMAEvent`` objects.
1665 Given the underlying data, and the start/stop points for each event,
1666 create the TMAEvent objects for the dayObs.
1668 Parameters
1669 ----------
1670 states : `list` of `tuple`
1671 The parsed states, as a list of tuples of the form:
1672 ``(eventStart, eventEnd, eventType, endReason)``
1673 dayObs : `int`
1674 The dayObs for the data.
1675 data : `pd.DataFrame`
1676 The merged dataframe.
1678 Returns
1679 -------
1680 events : `list` of `lsst.summit.utils.tmaUtils.TMAEvent`
1681 The events for the specified dayObs.
1682 """
1683 seqNum = 0
1684 events = []
1685 for parsedState in states:
1686 begin = data.iloc[parsedState.eventStart]["private_efdStamp"]
1687 end = data.iloc[parsedState.eventEnd]["private_efdStamp"]
1688 beginAstropy = efdTimestampToAstropy(begin)
1689 endAstropy = efdTimestampToAstropy(end)
1690 duration = end - begin
1691 event = TMAEvent(
1692 dayObs=dayObs,
1693 seqNum=seqNum,
1694 type=parsedState.previousState,
1695 endReason=parsedState.state,
1696 duration=duration,
1697 begin=beginAstropy,
1698 end=endAstropy,
1699 blockInfos=[], # this is added later
1700 _startRow=parsedState.eventStart,
1701 _endRow=parsedState.eventEnd,
1702 )
1703 events.append(event)
1704 seqNum += 1
1705 return events
1707 @staticmethod
1708 def printTmaDetailedState(tma):
1709 """Print the full state of all the components of the TMA.
1711 Currently this is the azimuth and elevation axes' power and motion
1712 states, and their respective inPosition statuses.
1714 Parameters
1715 ----------
1716 tma : `lsst.summit.utils.tmaUtils.TMAStateMachine`
1717 The TMA state machine in the state we want to print.
1718 """
1719 axes = ["azimuth", "elevation"]
1720 p = tma._parts
1721 axisPad = len(max(axes, key=len)) # length of the longest axis string == 9 here, but this is general
1722 motionPad = max(len(s.name) for s in AxisMotionState)
1723 powerPad = max(len(s.name) for s in PowerState)
1725 # example output to show what's being done with the padding:
1726 # azimuth - Power: ON Motion: STOPPED InPosition: True # noqa: W505
1727 # elevation - Power: ON Motion: MOVING_POINT_TO_POINT InPosition: False # noqa: W505
1728 for axis in axes:
1729 print(
1730 f"{axis:>{axisPad}} - "
1731 f"Power: {p[f'{axis}SystemState'].name:>{powerPad}} "
1732 f"Motion: {p[f'{axis}MotionState'].name:>{motionPad}} "
1733 f"InPosition: {p[f'{axis}InPosition']}"
1734 )
1735 print(f"Overall system state: {tma.state.name}")
1737 def printFullDayStateEvolution(self, dayObs, taiOrUtc="utc"):
1738 """Print the full TMA state evolution for the specified dayObs.
1740 Replays all the data from the EFD for the specified dayObs through
1741 the TMA state machine, and prints both the overall and detailed state
1742 of the TMA for each row.
1744 Parameters
1745 ----------
1746 dayObs : `int`
1747 The dayObs for which to print the state evolution.
1748 taiOrUtc : `str`, optional
1749 Whether to print the timestamps in TAI or UTC. Default is UTC.
1750 """
1751 # create a fake event which spans the whole day, and then use
1752 # printEventDetails code while skipping the header to print the
1753 # evolution.
1754 _ = self.getEvents(dayObs) # ensure the data has been retrieved from the EFD
1755 data = self._data[dayObs]
1756 lastRowNum = len(data) - 1
1758 fakeEvent = TMAEvent(
1759 dayObs=dayObs,
1760 seqNum=-1, # anything will do
1761 type=TMAState.OFF, # anything will do
1762 endReason=TMAState.OFF, # anything will do
1763 duration=-1, # anything will do
1764 begin=efdTimestampToAstropy(data.iloc[0]["private_efdStamp"]),
1765 end=efdTimestampToAstropy(data.iloc[-1]["private_efdStamp"]),
1766 _startRow=0,
1767 _endRow=lastRowNum,
1768 )
1769 self.printEventDetails(fakeEvent, taiOrUtc=taiOrUtc, printHeader=False)
1771 def printEventDetails(self, event, taiOrUtc="tai", printHeader=True):
1772 """Print a detailed breakdown of all state transitions during an event.
1774 Note: this is not the most efficient way to do this, but it is much the
1775 cleanest with respect to the actual state machine application and event
1776 generation code, and is easily fast enough for the cases it will be
1777 used for. It is not worth complicating the normal state machine logic
1778 to try to use this code.
1780 Parameters
1781 ----------
1782 event : `lsst.summit.utils.tmaUtils.TMAEvent`
1783 The event to display the details of.
1784 taiOrUtc : `str`, optional
1785 Whether to display time strings in TAI or UTC. Defaults to TAI.
1786 Case insensitive.
1787 printHeader : `bool`, optional
1788 Whether to print the event summary. Defaults to True. The primary
1789 reason for the existence of this option is so that this same
1790 printing function can be used to show the evolution of a whole day
1791 by supplying a fake event which spans the whole day, but this event
1792 necessarily has a meaningless summary, and so needs suppressing.
1793 """
1794 taiOrUtc = taiOrUtc.lower()
1795 if taiOrUtc not in ["tai", "utc"]:
1796 raise ValueError(f"Got unsuppoted value for {taiOrUtc=}")
1797 useUtc = taiOrUtc == "utc"
1799 if printHeader:
1800 print(
1801 f"Details for {event.duration:.2f}s {event.type.name} event dayObs={event.dayObs}"
1802 f" seqNum={event.seqNum}:"
1803 )
1804 print(f"- Event began at: {event.begin.utc.isot if useUtc else event.begin.isot}")
1805 print(f"- Event ended at: {event.end.utc.isot if useUtc else event.end.isot}")
1807 dayObs = event.dayObs
1808 data = self._data[dayObs]
1809 startRow = event._startRow
1810 endRow = event._endRow
1811 nRowsToApply = endRow - startRow + 1
1812 print(f"\nTotal number of rows in the merged dataframe: {len(data)}")
1813 if printHeader:
1814 print(f"of which rows {startRow} to {endRow} (inclusive) relate to this event.")
1816 # reconstruct all the states
1817 tma = TMAStateMachine(engineeringMode=True)
1818 _initializeTma(tma)
1820 tmaStates = {}
1821 firstAppliedRow = True # flag to print a header on the first row that's applied
1822 for rowNum, row in data.iterrows(): # must replay rows right from start to get full correct state
1823 if rowNum == startRow:
1824 # we've not yet applied this row, so this is the state just
1825 # before event
1826 print(f"\nBefore the event the TMA was in state {tma.state.name}:")
1827 self.printTmaDetailedState(tma)
1829 if rowNum >= startRow and rowNum <= endRow:
1830 if firstAppliedRow: # only print this intro on the first row we're applying
1831 print(
1832 f"\nThen, applying the {nRowsToApply} rows of data for this event, the state"
1833 " evolved as follows:\n"
1834 )
1835 firstAppliedRow = False
1837 # break the row down and print its details
1838 rowFor = row["rowFor"]
1839 axis, rowType = getAxisAndType(rowFor) # e.g. elevation, MotionState
1840 value = tma._getRowPayload(row, rowType, rowFor)
1841 valueStr = f"{str(value) if isinstance(value, bool) else value.name}"
1842 rowTime = efdTimestampToAstropy(row["private_efdStamp"])
1843 print(
1844 f"On row {rowNum} the {axis} axis had the {rowType} set to {valueStr} at"
1845 f" {rowTime.utc.isot if useUtc else rowTime.isot}"
1846 )
1848 # then apply it as usual, printing the state right afterwards
1849 tma.apply(row)
1850 tmaStates[rowNum] = tma.state
1851 self.printTmaDetailedState(tma)
1852 print()
1854 else:
1855 # if it's not in the range of interest then just apply it
1856 # silently as usual
1857 tma.apply(row)
1858 tmaStates[rowNum] = tma.state
1860 def findEvent(self, time):
1861 """Find the event which contains the specified time.
1863 If the specified time lies within an event, that event is returned. If
1864 it is at the exact start, that is logged, and if that start point is
1865 shared by the end of the previous event, that is logged too. If the
1866 event lies between events, the events either side are logged, but
1867 ``None`` is returned. If the time lies before the first event of the
1868 day a warning is logged, as for times after the last event of the day.
1870 Parameters
1871 ----------
1872 time : `astropy.time.Time`
1873 The time.
1875 Returns
1876 -------
1877 event : `lsst.summit.utils.tmaUtils.TMAEvent` or `None`
1878 The event which contains the specified time, or ``None`` if the
1879 time doesn't fall during an event.
1880 """
1881 # there are five possible cases:
1882 # 1) the time lies before the first event of the day
1883 # 2) the time lies after the last event of the day
1884 # 3) the time lies within an event
1885 # 3a) the time is exactly at the start of an event
1886 # 3b) if so, time can be shared by the end of the previous event if
1887 # they are contiguous
1888 # 4) the time lies between two events
1889 # 5) the time is exactly at end of the last event of the day. This is
1890 # an issue because event end times are exclusive, so this time is
1891 # not technically in that event, it's the moment it closes (and if
1892 # there *was* an event which followed contiguously, it would be in
1893 # that event instead, which is what motivates this definition of
1894 # lies within what event)
1896 dayObs = getDayObsForTime(time)
1897 # we know this is on the right day, and definitely before the specified
1898 # time, but sanity check this before continuing as this needs to be
1899 # true for this to give the correct answer
1900 assert getDayObsStartTime(dayObs) <= time
1901 assert getDayObsEndTime(dayObs) > time
1903 # command start to many log messages so define once here
1904 logStart = f"Specified time {time.isot} falls on {dayObs=}"
1906 events = self.getEvents(dayObs)
1907 if len(events) == 0:
1908 self.log.warning(f"There are no events found for {dayObs}")
1909 return None
1911 # check case 1)
1912 if time < events[0].begin:
1913 self.log.warning(f"{logStart} and is before the first event of the day")
1914 return None
1916 # check case 2)
1917 if time > events[-1].end:
1918 self.log.warning(f"{logStart} and is after the last event of the day")
1919 return None
1921 # check case 5)
1922 if time == events[-1].end:
1923 self.log.warning(
1924 f"{logStart} and is exactly at the end of the last event of the day"
1925 f" (seqnum={events[-1].seqNum}). Because event intervals are half-open, this"
1926 " time does not technically lie in any event"
1927 )
1928 return None
1930 # we are now either in an event, or between events. Walk through the
1931 # events, and if the end of the event is after the specified time, then
1932 # we're either in it or past it, so check if we're in.
1933 for eventNum, event in enumerate(events):
1934 if event.end > time: # case 3) we are now into or past the right event
1935 # the event end encloses the time, so note the > and not >=,
1936 # this must be strictly greater, we check the overlap case
1937 # later
1938 if time >= event.begin: # we're fully inside the event, so return it.
1939 # 3a) before returning, check if we're exactly at the start
1940 # of the event, and if so, log it. Then 3b) also check if
1941 # we're at the exact end of the previous event, and if so,
1942 # log that too.
1943 if time == event.begin:
1944 self.log.info(f"{logStart} and is exactly at the start of event" f" {eventNum}")
1945 if eventNum == 0: # I think this is actually impossible, but check anyway
1946 return event # can't check the previous event so return here
1947 previousEvent = events[eventNum - 1]
1948 if previousEvent.end == time:
1949 self.log.info(
1950 "Previous event is contiguous, so this time is also at the exact"
1951 f" end of {eventNum - 1}"
1952 )
1953 return event
1954 else: # case 4)
1955 # the event end is past the time, but it's not inside the
1956 # event, so we're between events. Log which we're between
1957 # and return None
1958 previousEvent = events[eventNum - 1]
1959 timeAfterPrev = (time - previousEvent.end).to_datetime()
1960 naturalTimeAfterPrev = humanize.naturaldelta(timeAfterPrev, minimum_unit="MICROSECONDS")
1961 timeBeforeCurrent = (event.begin - time).to_datetime()
1962 naturalTimeBeforeCurrent = humanize.naturaldelta(
1963 timeBeforeCurrent, minimum_unit="MICROSECONDS"
1964 )
1965 self.log.info(
1966 f"{logStart} and lies"
1967 f" {naturalTimeAfterPrev} after the end of event {previousEvent.seqNum}"
1968 f" and {naturalTimeBeforeCurrent} before the start of event {event.seqNum}."
1969 )
1970 return None
1972 raise RuntimeError(
1973 "Event finding logic fundamentally failed, which should never happen - the code" " needs fixing"
1974 )