-- I've rewritten the entire NS framework to use a producer / consumer model for input -> map and from map -> reduce. This is allowing us to scale reasonably efficiently up to 4 threads (see figure). Future work on the nano scheduler will be itemized in a separate JIRA entry.
-- Restructured the NS code for clarity. Docs everywhere.
-- This is considered version 1.0
-Off by default; engine fork isolates new code paths from old code paths,
so no integration tests change yet
-Experimental implementation is currently BROKEN due to a serious issue
involving file spans. No one can/should use the experimental features
until I've patched this issue.
-There are temporarily two independent versions of LocusIteratorByState.
Anyone changing one version should port the change to the other (if possible),
and anyone adding unit tests for one version should add the same unit tests
for the other (again, if possible). This situation will hopefully be extremely
temporary, and last only until the experimental implementation is proven.
-- Separate updating cumulative traversal metrics from printing progress. There's now an updateCumulativeMetrics function and a printProgress() that only takes a current position
-- printProgress now soles relies on the time since the last progress to decide if it will print or not. No longer uses the number of cycles, since this isn't reliable in the case of nano scheduling
-- GenomeAnalysisEngine now maintains a pointer to the master cumulative metrics. getCumulativeMetrics never returns null, which was handled in some parts of the code but not others.
-- Update all of the traversals to use the new updateCumulativeMetrics, printProgress model
-- Added progress callback to nano scheduler. Every bufferSize elements this callback is invoked, allowing us to smoothly update the progress meter in the NanoScheduler
-- Rename MapFunction to NanoSchedulerMap and the same for reduce.
- VariantAnnotatorEngine changed to call genotype annotations even if pilups and allele -> likelihood mappings are not present. Current genotype annotations altered to check for null pilupes and null mappings.
-- Confirmed that reads spanning off the end of the chromosome don't cause an exception by adding integration test for a single read that starts 7 bases from the end of chromosome 1 and spans 90 bases or so off. Added pileup integration test to ensure this behavior continues to work
-- Yes, GenomeLoc.compareTo was broken. The compareTo function only considered the contig and start position, but not the stop, when comparing genome locs.
-- Updated GenomeLoc.compareTo function to account for stop. Updated GATK code where necessary to fix resulting problems that depended on this.
-- Added unit tests to ensure that hashcode, equals, and compareTo are all correct for GenomeLocs
-- Stateless objects are required for nano-scheduling. This means you can take the RefMetaDataTracker provided by ReadBasedReferenceOrderedView, store it way, get another from the same view, and the original one behaves the same.
-- Previous behavior was unnecessary and causes all sorts of problems with RODs for reads. The old implementation simply failed in this case. The new code handles this correctly by forcing shards to have all of their data on a single contig.
-- Added a PrintReads integration test to ensure this behavior is correct
-- Adding test BAMs that have < 200 reads and span across contig boundaries
-- Deleted ReadMetaDataTracker
-- Added function to ReadShard to give us the span from the left most position of the reads in the shard to the right most, which is needed for the new view
-- ReadMetaDataTracker is dead! Long live the RefMetaDataTracker. Read walkers will soon just take RefMetaDataTracker objects. In this commit they take a class that trivially extends them
-- Rewrote ReadBasedReferenceOrderedView to produce RefMetaDataTrackers not the old class.
-- This new implementation produces thread-safe objects (i.e., holds no points to shared state). Suitable for use (to be tested) with nano scheduling
-- Simplified interfaces to use the simplest data structures (PeekableIterator) not the LocusAwareSeekableIterator, since I both hate those classes and this is on the long term trajectory to remove those from the GATK entirely.
-- Massively expanded DataProvider unit tests for ReadBasedReferenceOrderedView
-- Note that the old implementation of offset -> ROD in ReadRefMetaDataTracker was broken for any read not completely matching the reference. Rather than provide broken code the ReadMetaDataTracker only provides a "bag of RODs" interface. If you want to work with the relationship between the read and the RODs in your tool you need to manage the CIGAR element itself.
-- This commit breaks the new read walker BQSR, but Ryan knows this is coming
-- Subsequent commit will be retiring / fixing ValidateRODForReads
-- Groups inputs for each thread so that we don't have one thread execution per map() call
-- Added shutdown function
-- Documentation everywhere
-- Code cleanup
-- Extensive unittests
-- At this point I'm ready to integrate it into the engine for CPU parallel read walkers
– Write general NanoScheduler framework in utils.threading. Test with reading via iterator from list of integers, map is int * 2, reduce is sum. Should be efficiency using resources to do sum of 2 * (sum(1 - X)).
Done!
CPU parallelism is nano threads. Pfor across read / map / reduce. Use work queue to implement.
Create general read map reduce framework in utils. Test parallelism independently before hooking up to Locus iterator
Represent explicitly the dependency graph. Scheduler should choose the work units that are ready for computation, that are marked as "completing a computation", and then finally that maximize the number of sequent available work units. May be worth measuring expected cost for read read / map / reduce unit and use it to balance the compute
As input is single threaded just need one thread to populate inputs, which runs as fast as possible on parallel pushing data to fixed size queue. Each push creates map job and links to upcoming reduce job.
Note that there's at most one thread for IO tasks, and all of the threads can contribute to CPU tasks
-- Invert logic in GATKArgumentCollection to disable monitoring, not enable. That means monitoring is on by default
-- Fix testing error in unit tests
-- Rename variables in ThreadAllocation to be clearer
-- Old version StateMonitoringThreadFactory refactored into base class ThreadEfficiencyMonitor and subclass EfficiencyMonitoringThreadFactory.
-- Base class is used by LinearMicroScheduler to monitor performance of GATK in single threaded mode
-- MicroScheduler now handles management of the efficiency monitor. Includes master thread in monitor, meaning that reduce is now included for both schedulers
-- Allows us to ID (by proxy) time spent doing IO
-- Refactor StateMonitoryingThreadFactory to use it's own enum, not Thread.State
-- Reliable unit tests across mac and unix
David Roazen
Menachem Fromer
Eric Banks
Mark DePristo
Christopher Hartl
Ryan Poplin
Guillermo del Angel