Version 18 (modified by mhuber@…, 3 months ago)

--

PSNSC_MOPS requirements and notes (draft) --

Processing products

czar needs to make sure all faults and issues are cleared daily (early morning) for OSS, MSS, ESS, ThreePi?? products

  • chip-cam-warp -- nightly
  • warp-stack diffims -- nightly
  • datastore of WS diffim catalogs -- nightly
  • stamps -- chip only from pstamp, MOPS has own stamp server that requires uncompressed image access
  • MOPS needs visit 1-2, 2-3 when visit 4 lost due to weather/faults to increase just a pair to triple for tracklets
    • lost visit mapping to be done -- typical losses from fault at camera for any visit due to FWHM cut, v4 lost often due to weather/closure
      v1 missing: v2-v3
      v2 missing: v1-v3
      v3 missing: v2-v4
      v4 missing: v2-v3
      
    • now nightly_science.pl attempts at end-of-night/after observing if DESPERATE_DIFFS true

* sometime seem to be auto-made and other times not? need update on both when known -- appears to be at least partially due to the .multi nightly science task (which is not right and shouldn't be happening as .multi was supposed to be multiple night return visits?) -- .multi look to be made intranight when the camera stage quality faults on visit 2 & 3 and then makes a visit 1-4 diffim since neither 1 nor 4 is used to make a diffim -- expect other combinations to then be possible as well?

NEO targeted observations

FSS.nightlyscience, FSS.WS -- also needs to be finished by early morning

  • obs_mode FSS (new 201608) -- 6 exposure, MD-like dither (rotation) plus small offsets -- so WSdiff only and published to normal MOPS datastore location
  • chip->warp normal stdscience pantasks as with xSS
  • WSdiff currently runs in ~ippmops/stdscience_ws to make uncompressed WSdiffs

Cleanup

Specific for MOPS needs

  • chip -- 1 days
  • warp -- ? days (no specific requirement)
  • WW, WSdiff -- 1 days -- if disk space issue, can reduce this to 2 days as long as warps uncleaned the update time is minimal
    • maybe longer for the FSS WSdiff since special follow-up field
  • datastore WSdiff catalog bundles -- Bill has script doing?
  • pstamp -- manual cleanup like other ps_ud_%

Tasks/Improvements/Enhancements needed

Nightly products -- only really need image+mask (uncompressed) so could make normal use of no weight file to save disk use during nightly

  • currently hacked to not produce mask+weight in ippitc due to overuse of itc datanodes with non-nightly products...

WW diffims for MOPS --

  • excess detections causing full rejection of OTA -- typically edges full of junk, unclear if could be improved with other cuts

centroids poor causing large GCR -- factor of 2 larger..

  • Rob did comparisons of ipp-20141024/ipp-20170121 and products -- appears to be due to issues in all WWdiffs, not problem in SMF, warp, WSdiff
  • will add checks with MOPS tests and single convolution direction

WS diffims for MOPS --

  • image files uncompressed -- either in processing or a copy on ipp032 -- ~10TB/night extra space used (20TB since data kept for 36hrs), processing seems ok (network i/o, dtime, faults were all acceptable)
  • update w-band template stacks -- see below
  • could be using WS for i-band always -- need to test detection threshold like done for WWdiff (test chunk is in w-band, but use same threshold for all filters in WWdiff and even that could/should be adjusted at least for dynamic s/n cuts in MOPS)
  • exposure uniqueness required -- update .mops catalog format? (see below)
  • publishing -- TBD

ppTranslate/ppMOPS -- .mops catalog format update/revision --

  • evaluate contents again, remove/add new columns as needed -- must add skycell_id, chip XY and position
  • merge WW and WSdiff catalog publishing to single product if possible -- conflict with timing possibly

w-band stack update for WSdiff --

  • best seeing, uniform pixel coverage, WS level depth (about 10-20 best inputs to possibly allow quad stacks as well)
  • warps ready, need to run stacking and evaluate coverage (cleanup of poor edge cases)
  • by default will use the different LAP.PV3W set RA~7-16hr unless find a problem

MOPS test samples --

  • daily MOPS test quad -- label MOPS.dailytestset -- MOPS.dailytestset -- still in development
  • ops tag modification test chunk -- MOPS.testchunk.OSSR.R15S2.0.Q.w.20140603
    • necessary comparison set for WW and WS diff for any development/modifications to ops tag
    • need to define a test chunk for i-band probably -- different s/n dynamics? different stacks in WSdiffs at the very least
  • special case test quads -- fast movers, comets, other? -- quads need to be selected/defined and run through IPP
  • more automated checking for missing visit WWdiffs -- create by exposure status table in czardb for a night reset @initday?

MOPS stamp server @ITC -- MOPS using own node for this now 8/2017

  • needs some raid1 disk (minimum few 100 GB, more in future for other work), modest RAM and cores -- more RAM+cores, more stamps can be made
  • must be a reliable node and close to the data nodes on the network (and out to neo cluster) -- unclear which is to be used
  • will copy off images for stamps -- can be compressed in future
  • scp/nfs/rcp debate for overheads

pstamp server backlog -- particularly w/ QUB stamps, in future may want to split pantasks so MOPS doesn't get behind large numbers of uploaded parse jobs