NEWS.md

In clustermq: Evaluate Function Calls on HPC Schedulers (LSF, SGE, SLURM, PBS/Torque)

clustermq 0.9.1

• Disconnect monitor (libzmq with -DZMQ_BUILD_DRAFT_API=1) is now optional (#317)
• Fix a bug where worker shutdown notifications can cause a crash (#306, #308, #310)
• Fix a bug where template values were not filled correctly (#309)
• Fix a bug where using Rf_error lead to improper cleanup of resources (#311)
• Fix a bug where maximum worker timeout was multiplied and led to undefined behavior
• Fix a bug where ZeroMQ's -Werror flag led to compilation issues on M1 Mac
• Fix a bug where SSH tests could error with timeout on high load
• Worker API: CMQMaster now needs to know add_pending_workers(n)
• Worker API: status report info now displays properly

clustermq 0.9.0

Features

• Reuse of common data is now supported (#154)
• Jobs now error instead of stalling upon unexpected worker disconnect (#150)
• Workers now error if they can not establish a connection within a time limit
• Error if n_jobs and max_calls_worker provide insufficient call slots (#258)
• Request 1 GB by default in SGE template (#298) @nickholway
• Error and warning summary now orders by index and severity (#304)
• A call can have multiple warnings forwarded, not only last

Bugfix

• Fix bug where max memory reporting by gc() may be in different column (#240)
• Fix passing numerical job_id to qdel in PBS (#265)
• The job port/id pool is now used properly upon binding failure (#270) @luwidmer
• Common data size warning is now only displayed when exceeding limits (#287)

Internal

• Complete rewrite of the worker API
• We no longer depend on the purrr package

clustermq 0.8.95

• We are now using ZeroMQ via Rcpp in preparation for v0.9 (#151)
• New multiprocess backend via callr instead of forking (#142, #197)
• Sending data on sockets is now blocking to avoid excess memory usage (#161)
• multicore, multiprocess schedulers now support logging (#169)
• New option clustermq.host can specify host IP or network interface name (#170)
• Template filling will now raise error for missing keys (#174, #198)
• Workers failing with large common data is improved (fixed?) (#146, #179, #191)
• Local connections are now routed via 127.0.0.1 instead of localhost (#192)
• Submit messages are different between local, multicore and HPC (#196)
• Functions exported by foreach now have their environment stripped (#200)
• Deprecation of log_worker=T/F argument is rescinded

clustermq 0.8.9

• New option clustermq.ssh.timeout for SSH proxy startup (#157) @brendanf
• New option clustermq.worker.timeout for delay before worker shutdown (#188)
• Fixed PBS/Torque docs, template and cleanup (#184, #186) @mstr3336
• Warning if common data is very large, set by clustermq.data.warning (#189)

clustermq 0.8.8

• Q, Q_rows have new arguments verbose (#111) and pkgs (#144)
• foreach backend now uses its dedicated API where possible (#143, #144)
• Number and size of objects common to all calls now work properly
• Templates are filled internally and no longer depend on infuser package

clustermq 0.8.7

• Q now has max_calls_worker argument to avoid walltime (#110)
• Submission messages now list size of common data (drake#800)
• All default templates now have an optional cores per job field (#123)
• foreach now treats .export (#124) and .combine (#126) correctly
• New option clustermq.error.timeout to not wait for clean shutdown (#134)
• SSH command is now specified via a template file (#122)
• SSH will now forward errors to the local process (#135)
• The Wiki is deprecated, use https://mschubert.github.io/clustermq/ instead

clustermq 0.8.6

• Progress bar is now shown before any workers start (#107)
• Socket connections are now authenticated using a session password (#125)
• Marked internal functions with @keywords internal
• Added vignettes for the User Guide and Technical Documentation

clustermq 0.8.5

• Added experimental support as parallel foreach backend (#83)
• Moved templates to package inst/ directory (#85)
• Added send_call to worker to evaluate arbitrary expressions (drake#501; #86)
• Option clustermq.scheduler is now respected if set after package load (#88)
• System interrupts are now handled correctly (rzmq#44; #73, #93, #97)
• Number of workers running/total is now shown in progress bar (#98)
• Unqualified (short) host names are now resolved by default (#104)

clustermq 0.8.4

• Fix error for qsys$reusable when using n_jobs=0/local processing (#75)
• Scheduler-specific templates are deprecated. Use clustermq.template instead
• Allow option clustermq.defaults to fill default template values (#71)
• Errors in worker processing are now shut down cleanly (#67)
• Progress bar now shows estimated time remaining (#66)
• Progress bar now also shown when processing locally
• Memory summary now adds estimated memory of R session (#69)

clustermq 0.8.3

• Support rettype for function calls where return type is known (#59)
• Reduce memory requirements by processing results when we receive them
• Fix a bug where cleanup, log_worker flag were not working for SGE/SLURM

clustermq 0.8.2

• Fix a bug where never-started jobs are not cleaned up
• Fix a bug where tests leave processes if port binding fails (#60)
• Multicore no longer prints worker debug messages (#61)

clustermq 0.8.1

• Fix performance issues for a high number of function calls (#56)
• Fix bug where multicore workers were not shut down properly (#58)
• Fix default templates for SGE, LSF and SLURM (misplaced quote)

clustermq 0.8.0

Features

• Templates changed: clustermq:::worker now takes only master as argument
• Creating workers is now separated from Q, enabling worker reuse (#45)
• Objects in the function environment must now be exported explicitly (#47)
• Added multicore qsys using the parallel package (#49)
• New function Q_rows using data.frame rows as iterated arguments (#43)
• Job summary will now report max memory as reported by gc (#18)

Bugfix

• Fix a bug where copies of common_data are collected by gc too slowly (#19)

Internal

• Messages on the master are now processed in threads (#42)
• Jobs will now be submitted as array if possible

clustermq 0.7.0

• Initial release on CRAN

clustermq documentation built on Nov. 21, 2023, 5:06 p.m.