GitLab

bugfix: the disconnect_when_inactive(0) is no longer needed, since it is the default coming from URLs -- thanks to Michael Nuhn for reporting

This is achieved by using executables that mimic LSF tools (bjobs, lsid,
etc)

* version/2.2:
  Better way of sharing the "take_time" parameter

* version/2.1:
  Better way of sharing the "take_time" parameter

* version/2.0:
  Better way of sharing the "take_time" parameter

* version/1.9:
  Better way of sharing the "take_time" parameter

* version/2.2:
  Made _add_together() a method so that we can die if input data are missing
  Renamed AnalysisStats::update() to update_stats_and_monitor() in order to restore the inherited update() method
  Added an option to accurately catch errors in pipes (needs bash to work)
  Use "bash -o pipefail" to make sure we catch errors on both sides of the pipe
  Explain that  is in minutes (but stored in secs)
  Set  (cause_of_death) to 'UNKNOWN' when not defined

* version/2.1:
  Made _add_together() a method so that we can die if input data are missing
  Renamed AnalysisStats::update() to update_stats_and_monitor() in order to restore the inherited update() method
  Added an option to accurately catch errors in pipes (needs bash to work)
  Use "bash -o pipefail" to make sure we catch errors on both sides of the pipe
  Explain that  is in minutes (but stored in secs)
  Set  (cause_of_death) to 'UNKNOWN' when not defined

* version/2.0:
  Made _add_together() a method so that we can die if input data are missing
  Renamed AnalysisStats::update() to update_stats_and_monitor() in order to restore the inherited update() method
  Added an option to accurately catch errors in pipes (needs bash to work)
  Use "bash -o pipefail" to make sure we catch errors on both sides of the pipe
  Explain that  is in minutes (but stored in secs)
  Set  (cause_of_death) to 'UNKNOWN' when not defined

* version/1.9:
  Made _add_together() a method so that we can die if input data are missing
  Renamed AnalysisStats::update() to update_stats_and_monitor() in order to restore the inherited update() method
  Added an option to accurately catch errors in pipes (needs bash to work)
  Use "bash -o pipefail" to make sure we catch errors on both sides of the pipe
  Explain that  is in minutes (but stored in secs)
  Set  (cause_of_death) to 'UNKNOWN' when not defined

Renamed AnalysisStats::update() to update_stats_and_monitor() in order to restore the inherited update() method

* version/2.2:
  The batch-size is not set by the workers, so they shouldn't update it
  Only update hive_capacity in analysis_stats when in DYNAMIC mode

* version/2.1:
  The batch-size is not set by the workers, so they shouldn't update it
  Only update hive_capacity in analysis_stats when in DYNAMIC mode

* version/2.0:
  The batch-size is not set by the workers, so they shouldn't update it
  Only update hive_capacity in analysis_stats when in DYNAMIC mode

* version/1.9:
  The batch-size is not set by the workers, so they shouldn't update it
  Only update hive_capacity in analysis_stats when in DYNAMIC mode

This way, we avoid overriding values set by an external source (a human,
for instance)

* version/2.2:
  Don't need %hash_interval, can simply use %hash_curr_workers
  bugfix: make sure the plot start exactly at $start_date
  Decreased the threshold for detecting rounding errors. This allows to show memory usage as low as 5Mb
  bugfix: alive workers are now displayed until the current time

cherry picked from commit master - bcd6efd7 and bugfixed
again to correctly initialize the first data-point