clustermq 0.9.6

• Large common data size is now reported correctly (#336)
• Template filling will no longer convert large numbers to scientific format
• Common data will no longer be duplicated when sending to workers

clustermq 0.9.5

• Fix a bug where an outdated system libzmq led to compilation errors (#327)
• New option clustermq.ports specifies eligible port range (#328) @michaelmayer2

clustermq 0.9.4

• Fix a bug where worker stats were shown as NA (#325) @luwidmer
• Worker API: env() now visibly lists environment if called without arguments

clustermq 0.9.3

• Fix a bug where BiocParallel did not export required objects (#302) @bhayete-empress
• Fix a bug where already finished workers were killed (#307) @luwidmer
• Fix a bug where worker results and stats could be garbage collected (#324) @kkman
• There is now an FAQ vignette with answers to frequently asked questions
• Worker API: send() now reports a call identifier that current() tracks

clustermq 0.9.2

• Fix a bug where SSH proxy would not cache data properly (#320) @wardfont
• Fix a bug where max_calls_worker was not respected (#322) @bhayete-empress
• Local parallelism (multicore, multiprocess) again uses local IP (#321) @wlandau
• Worker API: info() now also returns current worker and number of calls

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