status: Running
last update: 31 May 2012

This page is a record of the configuration and execution of the P202 reprocessing project, full reprocessing from DIGIs using Pass7 analysis code. This project involves reprocessing with Pass7 classification trees and new IRFs. This task will read DIGI files and emit RECON, MERIT, GCR and CAL ROOT files, and the standard array of FITS files. It will be a CPU-intensive and storage-intensive enterprise requiring months of elapsed time and of order 1 Pbyte of storage. At the time of this task beginning, there will be about 20,000 science runs in Fermi (3.5 years accumulation).

To avoid occupying a new 1 PB of disk space, the plan is to remove old RECON files once they have been reprocessed. This is a shell game that involves some amount of buffer space and then waiting until the new RECON file has been created and (to some extent) validated before removal. The old RECON files will be retained on tape in the HPSS system and they will be available via xroot (but with some delay as these large files are staged in).

The name "P202" derives from the word "processing" and the initial file version to be used for the output data products, e.g., r0123456789_v202_merit.root.

"New generation" tasks (using SCons, OO and common python scripts, etc.)

• P202-ROOT - This task reads DIGI and produces reprocessed RECON + CAL + GCR + MERIT + FILTEREDMERIT (photons) + ELECTRONMERIT

• P202-FITS - This task reads MERIT and produces FT1 (photons) + EXTENDEDFT1 + LS1 (merit-like FITS file for photons) + EXTENDEDLS1 + ELECTRONFITS file

  Datafile names, versions and locations

Data file version numbers for this reprocessing will begin with v202.

XROOT location and file naming

Location template:

/glast/Data/Flight/Reprocess/<reprocessName>/<dataType>

Locations for P202:

/glast/Data/Flight/Reprocess/P202/recon
/glast/Data/Flight/Reprocess/P202/cal
/glast/Data/Flight/Reprocess/P202/gcr
/glast/Data/Flight/Reprocess/P202/merit
/glast/Data/Flight/Reprocess/P202/filteredmerit
/glast/Data/Flight/Reprocess/P202/electronmerit
/glast/Data/Flight/Reprocess/P202/ft1
/glast/Data/Flight/Reprocess/P202/extendedft1
/glast/Data/Flight/Reprocess/P202/electronft1
/glast/Data/Flight/Reprocess/P202/ls1
/glast/Data/Flight/Reprocess/P202/extendedls1

File naming:

Note: 'procVer' is a field added to the file name (and the keyword "PROC_VER" in the primary header) added to the FFD 5/12/2010. Ref: http://fermi.gsfc.nasa.gov/ssc/dev/current_documents/Science_DP_FFD_RevA.pdf

Examples:

/glast/Data/Flight/Reprocess/P200/recon/r0239557414_v202_recon.root
/glast/Data/Flight/Reprocess/P200/cal/r0239557414_v202_cal.root
/glast/Data/Flight/Reprocess/P200/gcr/r0239557414_v202_gcr.root
/glast/Data/Flight/Reprocess/P200/merit/r0239557414_v202_merit.root
/glast/Data/Flight/Reprocess/P200/filteredmerit/r0239557414_v202_filteredmerit.root
/glast/Data/Flight/Reprocess/P200/electronmerit/r0239557414_v202_electronmerit.root
/glast/Data/Flight/Reprocess/P200/extendedft1/gll_xp_p202_r0239559565_v202.fit
/glast/Data/Flight/Reprocess/P200/ft1/gll_ph_p202_r0239559565_v202.fit
/glast/Data/Flight/Reprocess/P200/electronft1/gll_el_p202_r0239559565_v202.fit
/glast/Data/Flight/Reprocess/P200/extendedls1/gll_xe_p202_r0239559565_v202.fit
/glast/Data/Flight/Reprocess/P200/ls1/gll_ev_p202_r0239559565_v202.fit

DataCatalog location and naming

Logical directory and group template:

Data/Flight/Reprocess/<reprocessName>:<dataType>

Note that the <dataType> field (following the colon) is a DataCatalog 'group' name, and file names are of the form r<run#>.

Naming examples:

Data/Flight/Reprocess/P202:RECON r0239557414
Data/Flight/Reprocess/P202:CAL r0239557414
Data/Flight/Reprocess/P202:GCR r0239557414
Data/Flight/Reprocess/P202:MERIT r0239557414
Data/Flight/Reprocess/P202:FILTEREDMERIT r0239557414
Data/Flight/Reprocess/P202:EXTENDEDFT1 r0239557414
Data/Flight/Reprocess/P202:FT1 r0239557414
Data/Flight/Reprocess/P202:ELECTRONFT1 r0239557414
Data/Flight/Reprocess/P202:EXTENDEDLS1 r0239557414
Data/Flight/Reprocess/P202:LS1 r0239557414

Data Sample

The currently defined data sample for P202 reprocessing includes:

Generation of FITS files is a second step in the reprocessing and has only been run on the first year of data. Stay tuned...

NOTE: One run, 242429468, of type TrigTest was declared 'good for science' and has been included.

Bookkeeping

1. (This page): Define ingredients of reprocessing (processing code/configuration changes)
2. Processing History database: http://glast-ground.slac.stanford.edu/HistoryProcessing/HProcessingRuns.jsp?processingname=P202
   1. List of all reprocessings
   2. List of all data runs reprocessed
   3. Pointers to all input data files (-> dataCatalog)
   4. Pointers to associated task processes (-> Pipeline II status)
3. Data Catalog database: http://glast-ground.slac.stanford.edu/DataCatalog/folder.jsp
   1. Lists of and pointers to all output data files
   2. Meta data associated with each output data product

P202-ROOT

Status chronology

• 2/13/2012 - begin trials with final calibration and alignments from Leon; 5 runs reprocessed
• 2/14/2012 - trials continue with blocks of 15, 20, 25 and 50 runs reprocessed (each run generates ~20 batch jobs)
• 2/16/2012 - begin trickleStream production. Initial config:

  ===============================================================================
    TRICKLE PARMS
  ===============================================================================
  task =  P202-ROOT
  maxRuns =  19172
  firstStep =  setupRun
  steps =  [['/processRun processClump', 1500, 20], ['mergeClumps', 70, 1]]
  maxStreamsPerCycle =  20
  timePerCycle =  300
  ===============================================================================
  

• 2/21/2012 - One clump reprocessed with pointer to new mySQL DB (stream 710.0)
• 2/22/2012 - 776 runs complete. Pausing task.
• 3/15/2012 - resume task. New goal is 1-year of data (~5600 runs)
• 3/31/2012 - 1-year complete (5600 runs). There have been a few nasty problems which need to be fixed before continuing.

  S/W component	bug fix	status
  New ROOT version	5-min 'transaction timeout' triggered by xroot data server reboot	done 4/3/2012
  New GlastRelease	1) include new ROOT version (above); 2) exit with non-zero RC on ROOT write error	done 4/5/2012, GR 17-35-24-rp04
  New GPL_TOOLS	check size/checksum of file written to xroot with known size/checksum	pending
  Tuned xroot on new Dell servers	silent file truncation when volume fills up JIRA	done 4/4/2012 (100 MB min space limit -> 100 GB; file system space check cadence changed from 10 min to 2 min)
  New xroot client tools	complain when xroot data server fails on write	done 4/3/2012, v3.1.1
  New TSkim	1) new ROOT version (above); 2) complain on ROOT write errors	done 4/5/2012, v08-02-01
  New xroot redirector	required step toward enabling HPSS staging	done 4/3/2012, v3.1.1

  Note also that the FILTEREDMERIT files contain 42 more events than the EXTENDEDFT1 files; they should be identical.
• 4/5/2012 - resume task. New goal is entire science dataset.
• 4/10/2012 - Unknown 'glitch' may have caused a few 100's of jobs to crash and take sulky46 along with them.
• 4/11/2012 - 10:40pm lightening strikes SLAC power lines. Site-wide power outage. Stream 7795 was the last stream submitted prior to the outage.
• 4/12/2012 - due to possible overload of sulky46/u18 writing a lot of core files, have introduced one change to processClumps.py: prepend "ulimit -c 0;" to gleam command to disable all core file generation. This starts approx with run 7605 (+/-).
• 5/9/2012 - major pipeline issue...shut down pipeline and allow to drain (due to tomorrow's major outage)
• 5/10/2012 - 13:40 outage over.
  • Update GR from 17-35-24-rp04 to 17-35-24-rp07 in which the only change is replacing the 5-minute xroot time-out with 8 hours. This change effective with stream 14314 and previously failed pieces of four other runs: 14247.6, 14273.23, 14274.8, 14231.9.
  • Leon advises that as of today, calibrations are valid only thru ~15 Dec 2011 (run 345574915) - which is somewhere around stream 18,400. He asks Sasha to produce more up-to-date calibs.
• 5/18/2012 - all calibrations now valid through 6 May 2012. No need to pause P202 task.
• 5/28/2012 - 15:30 Complete (through 31 March 2012)
  • Data Catalog summary:  

    Name	Type	Files	Events	Size	Created (UTC)	Links
    CAL	Group	20229	44,125,599,595	128.7 TB	25-Jan-2012 00:53:31	Files
    ELECTRONMERIT	Group	20229	90,904,582	205.7 GB	25-Jan-2012 00:53:32	Files
    FILTEREDMERIT	Group	20229	6,291,396,710	5.3 TB	25-Jan-2012 00:53:29	Files
    GCR	Group	20229	44,123,014,456	942.7 GB	25-Jan-2012 00:53:31	Files
    MERIT	Group	20229	44,125,679,961	35.4 TB	25-Jan-2012 00:53:30	Files
    RECON	Group	20229	44,123,612,977	590.0 TB	25-Jan-2012 00:53:33	Files

    There are discrepancies to track down!
    Turns out to be three problematic runs/streams:
    • 272707024/5723 - I/O prob, corrupt files, entire stream rolled back
    • 279108810/6847 - xroot transient access prob., re-registered in dataCat
    • 284813327/7848 - xroot transient access prob., re-registered in dataCat
• Final trickleStream configuration:

  ===============================================================================
    TRICKLE PARMS
  ===============================================================================
  task =  P202-ROOT
  maxRuns =  20229
  firstStep =  setupRun
  steps =  [['/processRun processClump', 2000, 21], ['mergeClumps', 200, 1]]
  maxStreamsPerCycle =  20
  timePerCycle =  300
  ------DEBUG----------------
  maxCycles =  0
  chatter =  False
  dryRun =  False
  ===============================================================================
  

• 5/31/2012 - Rolling back all or part of the three runs above solved the discrepancies in # events.  New dataCatalog tally looks like this:

Configuration

Timing and Scaling

• processClump
  • with 1300 jobs completed, the average time to run varies by processor type from 220 min (hequ) to 370 min (boer).
    
  • with nearly 10,000 runs complete, the plots appear below:
    
• mergeClumps
  • with 42 jobs completed, the average time to run varies by processor type from 5-30 minutes.
    

Load balancing

trickleStream parameters:

P202-FITS

This task generates all desired FITS data products.

Status chronology

• 3/2/2012 - Define block 1 as the 776 runs in P202-ROOT block 1. Configure trickleStream and begin (14:08)
• 3/31/2012 - Define block 2 as 5600 runs. Reconfig trickleStream and begin (18:05)
• 4/01/2012 - Block 2 complete (most of the 4824 jobs completed in about six hours w/1000 job limit).

Configuration

Generation of output data products:

[1] /afs/slac/g/glast/ground/releases/volume04/evtClassDefs/00-19-04/data

Note that diffuse response is calculated for 'source' and 'clean' event classes only.

Note on 'Code Variant': The SLAC batch farm contains a mixture of architectures , both hardware (Intel/AMD 64-bit) and software (RHEL5-64, gcc v4.1, etc.). At this time, GlastRelease builds only on RHEL4-32, while ScienceTools builds for RHEL5-32, RHEL5-64.

Timing