rbbt-util 5.32.4 → 5.32.10

data/share/rbbt_commands/lsf/orchestrate

@@ -0,0 +1,58 @@
+#!/usr/bin/env ruby
+
+require 'rbbt/util/simpleopt'
+require 'rbbt/workflow'
+require 'rbbt/workflow/usage'
+require 'rbbt/hpc'
+require 'rbbt/hpc/orchestrate'
+require 'time'
+
+$slurm_options = SOPT.get <<EOF
+-dr--dry_run Print only the template
+-cj--clean_job Clean job
+--drbbt* Use development version of rbbt 
+-sing--singularity Use Singularity
+-si--singularity_img* Singularity image to use
+-ug--user_group* Use alternative user group for group project directory
+-c--contain* Contain in directory (using Singularity)
+-s--sync* Contain in directory and sync jobs
+-e--exclusive Make exclusive use of the node
+-hm--highmem Make use of highmem cores
+-wc--wipe_container* Wipe the jobs from the contain directory
+-CS--contain_and_sync Contain and sync to default locations
+-ci--copy_image When using a container directory, copy image there
+-t--tail Tail the logs
+-BPP--batch_procpath* Save Procpath performance for batch job; specify only options
+-q--queue* Queue
+-t--task_cpus* Tasks
+-tm--time* Time
+-lin--licenses* SLURM licenses
+-cons--constraint* SLURM constraint
+-W--workflows* Additional workflows
+-OR--orchestration_rules* Orchestration rules
+-rmb--remove_batch_basedir Remove the SLURM working directory (command, STDIN, exit status, ...)
+EOF
+
+batch_system = $slurm_options.delete :batch_system
+batch_system ||= 'auto'
+
+HPC::BATCH_MODULE = HPC.batch_system batch_system
+
+raise ParameterException.new("Could not detect batch_system: #{Misc.fingerprint batch_system}") if HPC::BATCH_MODULE.nil?
+
+class Step
+    def run(*args)
+        if done?
+            self.load
+        else
+            begin
+                Log.debug "Issuing SLURM job for #{self.path}"
+                HPC::BATCH_MODULE.orchestrate_job(self, SOPT::GOT_OPTIONS.merge($slurm_options)) 
+            rescue HPC::SBATCH
+            end
+        end
+    end
+end
+
+ARGV.concat ["-W", $slurm_options[:workflows], '--detach'] if $slurm_options[:workflows]
+load Rbbt.share.rbbt_commands.workflow.task.find

data/share/rbbt_commands/lsf/tail

@@ -0,0 +1,55 @@
+#!/usr/bin/env ruby
+
+require 'rbbt-util'
+require 'rbbt/util/simpleopt'
+require 'rbbt/hpc'
+
+#$0 = "rbbt #{$previous_commands*""} #{ File.basename(__FILE__) }" if $previous_commands
+
+options = SOPT.setup <<EOF
+
+Queue a job in Marenostrum
+
+$ rbbt slurm tail <directory> [options] 
+
+-h--help Print this help
+-d--done Done jobs only
+-e--error Error jobs only
+-a--aborted SLURM aboted jobs
+-r--running Running jobs only
+-q--queued Queued jobs only
+-j--job* Job ids
+-s--search* Regular expression
+-t--tail* Show the last lines of the STDERR
+-p--progress Report progress of job and the dependencies
+-SBP--sbatch_parameters show sbatch parameters
+-PERF--procpath_performance show Procpath performance summary
+-sacct--sacct_peformance show sacct performance summary
+-bs--batch_system* Batch system to use: auto, lsf, slurm (default is auto-detect)
+EOF
+
+if options[:help]
+ if defined? rbbt_usage
+  rbbt_usage 
+ else
+  puts SOPT.doc
+ end
+ exit 0
+end
+
+batch_system = options.delete :batch_system
+batch_system ||= 'auto'
+
+HPC::BATCH_MODULE = HPC.batch_system batch_system
+
+raise ParameterException.new("Could not detect batch_system: #{Misc.fingerprint batch_system}") if HPC::BATCH_MODULE.nil?
+
+directory = ARGV.shift
+
+raise ParameterException if directory.nil?
+
+directory = File.dirname(directory) unless File.directory?(directory)
+
+require 'rbbt/hpc/slurm'
+
+HPC::BATCH_MODULE.follow_job directory, true

data/share/rbbt_commands/lsf/task

@@ -0,0 +1,57 @@
+#!/usr/bin/env ruby
+
+require 'rbbt/util/simpleopt'
+require 'rbbt/workflow'
+require 'rbbt/workflow/usage'
+require 'rbbt/hpc'
+require 'time'
+
+$slurm_options = SOPT.get <<EOF
+-dr--dry_run Print only the template
+-cj--clean_job Clean job
+--drbbt* Use development version of rbbt 
+-sing--singularity Use Singularity
+-si--singularity_img* Singularity image to use
+-ug--user_group* Use alternative user group for group project directory
+-c--contain* Contain in directory (using Singularity)
+-s--sync* Contain in directory and sync jobs
+-e--exclusive Make exclusive use of the node
+-hm--highmem Make use of highmem cores
+-wc--wipe_container* Wipe the jobs from the contain directory
+-CS--contain_and_sync Contain and sync to default locations
+-ci--copy_image When using a container directory, copy image there
+-t--tail Tail the logs
+-BPP--batch_procpath* Save Procpath performance for batch job; specify only options
+-q--queue* Queue
+-t--task_cpus* Tasks
+-tm--time* Time
+-lin--licenses* SLURM licenses
+-cons--constraint* SLURM constraint
+-W--workflows* Additional workflows
+-rmb--remove_batch_dir Remove the batch working directory (command, STDIN, exit status, ...)
+-bs--batch_system* Batch system to use: auto, lsf, slurm (default is auto-detect)
+EOF
+
+batch_system = $slurm_options.delete :batch_system
+batch_system ||= 'auto'
+
+HPC::BATCH_MODULE = HPC.batch_system batch_system
+
+raise ParameterException.new("Could not detect batch_system: #{Misc.fingerprint batch_system}") if HPC::BATCH_MODULE.nil?
+
+class Step
+    def run(*args)
+        if done?
+            self.load
+        else
+            begin
+                Log.debug "Issuing SLURM job for #{self.path}"
+                HPC::BATCH_MODULE.run_job(self, SOPT::GOT_OPTIONS.merge($slurm_options)) 
+            rescue HPC::SBATCH
+            end
+        end
+    end
+end
+
+ARGV.concat ["-W", $slurm_options[:workflows]] if $slurm_options[:workflows]
+load Rbbt.share.rbbt_commands.workflow.task.find

data/share/rbbt_commands/migrate

@@ -2,7 +2,7 @@
 
 require 'rbbt-util'
 require 'rbbt/util/simpleopt'
-require 'rbbt/workflow/remote_workflow'
+require 'rbbt/util/migrate'
 
 $0 = "rbbt #{$previous_commands*""} #{ File.basename(__FILE__) }" if $previous_commands
 
@@ -30,83 +30,10 @@ if options[:help]
   exit 0
 end
 
-#excludes = %w(.save .crap .source tmp filecache open-remote workflows apps software jobs PCAWG)
-excludes = %w(.save .crap .source tmp filecache open-remote)
-excludes += (options[:exclude] || "").split(/,\s*/)
-excludes_str = excludes.collect{|s| "--exclude '#{s}'" } * " "
-
-test_str = options[:test] ? '-nv' : ''
-
 path, search_path, _sep, *other = ARGV
 
 search_path = 'user' if search_path.nil?
-resource = Rbbt
-
-path, real_paths, lpath = if options[:source]
-          lpath, *paths = Misc.ssh_run(options[:source], <<-EOF).split("\n")
-require 'rbbt-util'
-path = "#{path}"
-if Open.exists?(path)
-  path = #{resource.to_s}.identify(path)
-else
-  path = Path.setup(path)
-end
-puts path
-puts path.glob_all.collect{|p| File.directory?(p) ? p + "/" : p } * "\n"
-          EOF
-          [path, paths.collect{|p| [options[:source], p] * ":"}, lpath]
-        else
-          if File.exists?(path)
-
-            path = resource.identify(path)
-          else
-            path = Path.setup(path)
-          end
-          [path, path.glob_all, path]
-        end
-
-target = if options[:target] 
-           target = Misc.ssh_run(options[:target], <<-EOF).split("\n").first
-require 'rbbt-util'
-path = "#{path}"
-resource = #{resource.to_s}
-search_path = "#{search_path}"
-puts resource[path].find(search_path)
-           EOF
-         else
-           resource[lpath].find(search_path)
-         end
-
-real_paths.each do |source|
-
-
-  if File.directory?(source) || source =~ /\/$/
-    source += "/" unless source[-1] == "/"
-    target += "/" unless target[-1] == "/"
-  end
 
-  next if source == target
+options[:other] = other
 
-  if options[:target]
-    CMD.cmd("ssh #{options[:target]} mkdir -p '#{File.dirname(target)}'")
-  else
-    Open.mkdir File.dirname(target)
-  end
-
-  if options[:target]
-   target_path = [options[:target], target] * ":" 
-  else
-   target_path = target
-  end
-
-  cmd = "rsync -avztAXHP --copy-unsafe-links #{test_str} #{excludes_str} #{source} #{target_path} #{other * " "}"
-
-  cmd << " && rm -Rf #{source}" if options[:delete]
-
-  if options[:print]
-    puts cmd 
-    exit 0
-  else
-    CMD.cmd_log(cmd, :log => Log::INFO)
-  end
-end
+Rbbt.migrate(path, search_path, options)

data/share/rbbt_commands/slurm/clean

@@ -0,0 +1,212 @@
+#!/usr/bin/env ruby
+
+require 'rbbt-util'
+require 'rbbt/util/simpleopt'
+require 'rbbt/hpc'
+
+#$0 = "rbbt #{$previous_commands*""} #{ File.basename(__FILE__) }" if $previous_commands
+
+options = SOPT.setup <<EOF
+
+Clean error or aborted jobs
+
+$ rbbt slurm clean [options] 
+
+-h--help Print this help
+-d--done Done jobs only
+-e--error Error jobs only
+-a--aborted SLURM aboted jobs
+-q--queued Queued jobs only
+-j--job* Job ids
+-s--search* Regular expression
+-t--tail* Show the last lines of the STDERR
+-BP--batch_parameters show batch parameters
+-dr--dry_run Do not erase anything
+EOF
+
+if options[:help]
+ if defined? rbbt_usage
+  rbbt_usage 
+ else
+  puts SOPT.doc
+ end
+ exit 0
+end
+
+batch_system = options.delete :batch_system
+batch_system ||= 'auto'
+
+HPC::BATCH_MODULE = HPC.batch_system batch_system
+
+raise ParameterException.new("Could not detect batch_system: #{Misc.fingerprint batch_system}") if HPC::BATCH_MODULE.nil?
+
+Log.severity = 4
+done, error, aborted, queued, jobid, search, tail, batch_parameters, dry_run = options.values_at :done, :error, :aborted, :queued, :job, :search, :tail, :batch_parameters, :dry_run
+
+workdir = File.expand_path('~/rbbt-batch')
+Path.setup(workdir)
+
+running_jobs = begin
+                squeue_txt = HPC::BATCH_MODULE.job_status
+                squeue_txt.split("\n").collect{|l| l.to_i.to_s}
+               rescue
+                Log.warn "Cannot determine if jobs are running, they will seem to be all alive (Job ID in green)"
+                squeue_txt = nil
+                $norunningjobs = true
+                []
+               end
+
+if squeue_txt
+ job_nodes = {}
+ squeue_txt.split("\n").each do |line|
+  parts = line.strip.split(/\s+/)
+  job_nodes[parts.first] = parts.last.split(",")
+ end
+else
+ job_nodes = nil
+end
+
+count = 0
+workdir.glob("**/command.batch").sort_by{|f| File.mtime(f)}.each do |fcmd|
+ dir = File.dirname(fcmd)
+ command_txt = Open.read(fcmd)
+
+ if m = command_txt.match(/#CMD: (.*)/)
+  cmd = m[1]
+ else
+  cmd = nil
+ end
+
+ if m = command_txt.match(/# Run command\n(.*?)\n/im)
+  exe = m[1]
+ else
+  exe = nil
+ end
+
+ if m = command_txt.match(/^CONTAINER_DIR=(.*)/)
+  container_home = m[1]
+ else
+  container_home = nil
+ end
+
+ if m = command_txt.match(/^BATCH_SYSTEM=(.*)/)
+  job_batch_system = m[1].downcase
+ else
+  job_batch_system = nil
+ end
+
+ different_system = job_batch_system != batch_system
+
+ if File.exists?(fid = File.join(dir, 'job.id'))
+  id = Open.read(fid).chomp
+ else
+  id = nil
+ end
+
+ if File.exists?(fstatus = File.join(dir, 'exit.status'))
+  exit_status = Open.read(fstatus).to_i
+ else
+  exit_status = nil
+ end
+
+ if File.exists?(fstatus = File.join(dir, 'job.status'))
+   fstatus_txt = Open.read(fstatus) 
+   begin
+     if job_batch_system == "lsf"
+       nodes = Open.read(fstatus).split("\n").last.split(/\s+/)[5].split(",")
+     else
+       nodes = Open.read(fstatus).split("\n").last.split(/\s+/).last.split(",")
+     end
+   rescue
+     nodes = []
+   end
+ elsif job_nodes[id]
+  nodes = job_nodes[id]
+ else
+  nodes = []
+ end
+
+ if File.exists?(File.join(dir, 'std.out'))
+  outt = File.mtime File.join(dir, 'std.out')
+  errt = File.mtime File.join(dir, 'std.err')
+  time_diff = Time.now - [outt, errt].max
+ end
+
+ fdep = File.join(dir, 'dependencies.list')
+ deps = Open.read(fdep).split("\n") if File.exists?(fdep)
+
+ fcadep = File.join(dir, 'canfail_dependencies.list')
+ cadeps = Open.read(fcadep).split("\n") if File.exists?(fcadep)
+
+ aborted = error = true if aborted.nil? && error.nil?
+ #if done || error || aborted || running || queued || jobid || search
+ #  select = false
+ #  select = true if done && exit_status && exit_status.to_i == 0 
+ #  select = true if error && exit_status && exit_status.to_i != 0 
+ #  select = true if aborted && (exit_status.nil? && ! running_jobs.include?(id))
+ #  select = select && jobid.split(",").include?(id) if jobid 
+ #  select = select && cmd.match(/#{search}/) if search
+ #  next unless select 
+ #end
+
+ if done || error || aborted || queued || jobid 
+   select = false
+   select = true if done && exit_status == 0
+   select = true if error && exit_status && exit_status != 0 
+   select = true if aborted && (exit_status.nil? && ! running_jobs.include?(id))
+   is_running =  exit_status.nil? && ( (running_jobs.include?(id) && (!deps || (running_jobs & deps).empty?)) || different_system )
+   select = true if queued && deps && (running_jobs & deps).any? || queued && is_running && nodes.empty?
+   select = true if jobid && jobid.split(",").include?(id)
+   select = select && cmd.match(/#{search}/) if search
+   next unless select 
+ elsif search
+   select = false
+   select = true if search && cmd.match(/#{search}/)
+   next unless select 
+ end
+
+
+ puts Log.color(:yellow, "**ERASING**")
+ puts Log.color :blue, dir
+ puts Log.color(:magenta, "Creation: ") << File.mtime(File.join(dir, 'command.batch')).to_s
+ puts Log.color(:magenta, "Done: ") << File.mtime(File.join(dir, 'exit.status')).to_s if File.exist?(File.join(dir, 'exit.status'))
+ puts Log.color(:magenta, "Exec: ") << (exe || "Missing")
+ puts Log.color(:magenta, "CMD: ") << (Log.color(:yellow, cmd) || "Missing")
+ puts Log.color(:magenta, "HOME: ") << Log.color(:yellow, container_home) if container_home
+ puts Log.color(:magenta, "Job ID: ") << (exit_status ? (exit_status == 0 ? Log.color(:green, "Done") : Log.color(:red, "Error")) + " (#{ id })" :  (running_jobs.include?(id) || $norunningjobs ? Log.color(:green, id) : Log.color(:red, id) ))
+ puts Log.color(:magenta, "Dependencies: ") << deps * ", " if deps
+ puts Log.color(:magenta, "Dependencies (can fail): ") << cadeps * ", " if cadeps
+ puts Log.color(:magenta, "Nodes: ") << nodes * ", "
+ puts Log.color(:magenta, "Output: ") << File.exists?(File.join(dir, 'std.out')).to_s << (id.nil? ? "" : " (last update " + Misc.format_seconds(time_diff) + " ago)")
+
+ if options[:batch_parameters]
+   puts Log.color(:magenta, "BATCH parameters: ")
+   case job_batch_system 
+   when 'slurm'
+     puts Log.color :blue, CMD.cmd('grep "^#SBATCH" |tail -n +6', :in => Open.read(fcmd)).read.strip
+   when 'lsf'
+     puts Log.color :blue, CMD.cmd('grep "^#BSUB" |tail -n +6', :in => Open.read(fcmd)).read.strip
+   end
+ end
+
+ if tail && File.exists?(File.join(dir, 'std.err'))
+   if exit_status && exit_status != 0
+     puts Log.color(:magenta, "First error or exception found: ")
+     puts CMD.cmd("grep -i -w 'error\\|[a-z]*exception' #{File.join(dir, 'std.err')} -A #{tail.to_i} |head -n #{tail.to_i}", :no_fail => true).read
+   elsif exit_status
+     puts Log.color(:magenta, "Completed jobs: ")
+     puts CMD.cmd("grep -i -w 'Completed step' #{File.join(dir, 'std.err')} | grep -v 'Retrying dep.' | tail -n #{tail.to_i}", :no_fail => true).read
+   else
+     puts Log.color(:magenta, "Log tail: ")
+     puts CMD.cmd("tail -n #{tail.to_i} #{File.join(dir, 'std.err')}").read
+   end
+ end
+
+ count += 1
+
+ Open.rm_rf dir unless dry_run
+end
+
+puts 
+puts "Found #{count} jobs"
+