Added ForceRecord() method to enable writing remaining log changes before exiting.
[arvados.git] / crunch_scripts / run-command
index 9e8c13fcc6b52c0d268d7f9015c3038bdbe31712..c1e747506091863a11b028656b662e50f484ef5b 100755 (executable)
@@ -1,5 +1,13 @@
 #!/usr/bin/env python
 
+import logging
+
+logger = logging.getLogger('run-command')
+log_handler = logging.StreamHandler()
+log_handler.setFormatter(logging.Formatter("run-command: %(message)s"))
+logger.addHandler(log_handler)
+logger.setLevel(logging.INFO)
+
 import arvados
 import re
 import os
@@ -15,29 +23,42 @@ import copy
 import traceback
 import pprint
 import multiprocessing
-import logging
 import crunchutil.robust_put as robust_put
 import crunchutil.vwd as vwd
+import argparse
+import json
+import tempfile
 
-os.umask(0077)
-logging.basicConfig(format="run-command: %(message)s")
-
-t = arvados.current_task().tmpdir
+parser = argparse.ArgumentParser()
+parser.add_argument('--dry-run', action='store_true')
+parser.add_argument('--script-parameters', type=str, default="{}")
+args = parser.parse_args()
 
-api = arvados.api('v1')
+os.umask(0077)
 
-os.chdir(arvados.current_task().tmpdir)
-os.mkdir("tmpdir")
-os.mkdir("output")
+if not args.dry_run:
+    api = arvados.api('v1')
+    t = arvados.current_task().tmpdir
+    os.chdir(arvados.current_task().tmpdir)
+    os.mkdir("tmpdir")
+    os.mkdir("output")
 
-os.chdir("output")
+    os.chdir("output")
 
-outdir = os.getcwd()
+    outdir = os.getcwd()
 
-taskp = None
-jobp = arvados.current_job()['script_parameters']
-if len(arvados.current_task()['parameters']) > 0:
-    taskp = arvados.current_task()['parameters']
+    taskp = None
+    jobp = arvados.current_job()['script_parameters']
+    if len(arvados.current_task()['parameters']) > 0:
+        taskp = arvados.current_task()['parameters']
+else:
+    outdir = "/tmp"
+    jobp = json.loads(args.job_parameters)
+    os.environ['JOB_UUID'] = 'zzzzz-8i9sb-1234567890abcde'
+    os.environ['TASK_UUID'] = 'zzzzz-ot0gb-1234567890abcde'
+    os.environ['CRUNCH_SRC'] = '/tmp/crunche-src'
+    if 'TASK_KEEPMOUNT' not in os.environ:
+        os.environ['TASK_KEEPMOUNT'] = '/keep'
 
 links = []
 
@@ -74,6 +95,12 @@ class SigHandler(object):
         sp.send_signal(signum)
         self.sig = signum
 
+def add_to_group(gr, match):
+    m = match.groups()
+    if m not in gr:
+        gr[m] = []
+    gr[m].append(match.group(0))
+
 def expand_item(p, c):
     if isinstance(c, dict):
         if "foreach" in c and "command" in c:
@@ -85,17 +112,50 @@ def expand_item(p, c):
                 params[var] = i
                 r.extend(expand_list(params, c["command"]))
             return r
+        if "list" in c and "index" in c and "command" in c:
+            var = c["list"]
+            items = get_items(p, p[var])
+            params = copy.copy(p)
+            params[var] = items[int(c["index"])]
+            return expand_list(params, c["command"])
+        if "regex" in c:
+            pattern = re.compile(c["regex"])
+            if "filter" in c:
+                items = get_items(p, p[c["filter"]])
+                return [i for i in items if pattern.match(i)]
+            elif "group" in c:
+                items = get_items(p, p[c["group"]])
+                groups = {}
+                for i in items:
+                    match = pattern.match(i)
+                    if match:
+                        add_to_group(groups, match)
+                return [groups[k] for k in groups]
+            elif "extract" in c:
+                items = get_items(p, p[c["extract"]])
+                r = []
+                for i in items:
+                    match = pattern.match(i)
+                    if match:
+                        r.append(list(match.groups()))
+                return r
     elif isinstance(c, list):
         return expand_list(p, c)
-    elif isinstance(c, str) or isinstance(c, unicode):
+    elif isinstance(c, basestring):
         return [subst.do_substitution(p, c)]
 
     return []
 
 def expand_list(p, l):
-    return [exp for arg in l for exp in expand_item(p, arg)]
+    if isinstance(l, basestring):
+        return expand_item(p, l)
+    else:
+        return [exp for arg in l for exp in expand_item(p, arg)]
 
 def get_items(p, value):
+    if isinstance(value, dict):
+        return expand_item(p, value)
+
     if isinstance(value, list):
         return expand_list(p, value)
 
@@ -104,10 +164,10 @@ def get_items(p, value):
     prefix = fn[len(os.environ['TASK_KEEPMOUNT'])+1:]
     if mode is not None:
         if stat.S_ISDIR(mode):
-            items = ["$(dir %s/%s/)" % (prefix, l) for l in os.listdir(fn)]
+            items = [os.path.join(fn, l) for l in os.listdir(fn)]
         elif stat.S_ISREG(mode):
             with open(fn) as f:
-                items = [line for line in f]
+                items = [line.rstrip("\r\n") for line in f]
         return items
     else:
         return None
@@ -118,58 +178,87 @@ stdinname = None
 stdinfile = None
 rcode = 1
 
-try:
-    if "task.foreach" in jobp:
-        if arvados.current_task()['sequence'] == 0:
-            var = jobp["task.foreach"]
-            items = get_items(jobp, jobp[var])
-            logging.info("parallelizing on %s with items %s" % (var, items))
-            if items is not None:
-                for i in items:
-                    params = copy.copy(jobp)
-                    params[var] = i
+def recursive_foreach(params, fvars):
+    var = fvars[0]
+    fvars = fvars[1:]
+    items = get_items(params, params[var])
+    logger.info("parallelizing on %s with items %s" % (var, items))
+    if items is not None:
+        for i in items:
+            params = copy.copy(params)
+            params[var] = i
+            if len(fvars) > 0:
+                recursive_foreach(params, fvars)
+            else:
+                if not args.dry_run:
                     arvados.api().job_tasks().create(body={
                         'job_uuid': arvados.current_job()['uuid'],
                         'created_by_job_task_uuid': arvados.current_task()['uuid'],
                         'sequence': 1,
                         'parameters': params
-                        }
-                    ).execute()
+                    }).execute()
+                else:
+                    logger.info(expand_list(params, params["command"]))
+    else:
+        logger.error("parameter %s with value %s in task.foreach yielded no items" % (var, params[var]))
+        sys.exit(1)
+
+try:
+    if "task.foreach" in jobp:
+        if args.dry_run or arvados.current_task()['sequence'] == 0:
+            # This is the first task to start the other tasks and exit
+            fvars = jobp["task.foreach"]
+            if isinstance(fvars, basestring):
+                fvars = [fvars]
+            if not isinstance(fvars, list) or len(fvars) == 0:
+                logger.error("value of task.foreach must be a string or non-empty list")
+                sys.exit(1)
+            recursive_foreach(jobp, jobp["task.foreach"])
+            if not args.dry_run:
                 if "task.vwd" in jobp:
-                    # Base vwd collection will be merged with output fragments from
-                    # the other tasks by crunch.
+                    # Set output of the first task to the base vwd collection so it
+                    # will be merged with output fragments from the other tasks by
+                    # crunch.
                     arvados.current_task().set_output(subst.do_substitution(jobp, jobp["task.vwd"]))
                 else:
                     arvados.current_task().set_output(None)
-                sys.exit(0)
-            else:
-                sys.exit(1)
+            sys.exit(0)
     else:
+        # This is the only task so taskp/jobp are the same
         taskp = jobp
 
-    if "task.vwd" in taskp:
-        # Populate output directory with symlinks to files in collection
-        vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir)
+    if not args.dry_run:
+        if "task.vwd" in taskp:
+            # Populate output directory with symlinks to files in collection
+            vwd.checkout(subst.do_substitution(taskp, taskp["task.vwd"]), outdir)
 
-    if "task.cwd" in taskp:
-        os.chdir(subst.do_substitution(taskp, taskp["task.cwd"]))
+        if "task.cwd" in taskp:
+            os.chdir(subst.do_substitution(taskp, taskp["task.cwd"]))
 
     cmd = expand_list(taskp, taskp["command"])
 
-    if "task.stdin" in taskp:
-        stdinname = subst.do_substitution(taskp, taskp["task.stdin"])
-        stdinfile = open(stdinname, "rb")
+    if not args.dry_run:
+        if "task.stdin" in taskp:
+            stdinname = subst.do_substitution(taskp, taskp["task.stdin"])
+            stdinfile = open(stdinname, "rb")
 
-    if "task.stdout" in taskp:
-        stdoutname = subst.do_substitution(taskp, taskp["task.stdout"])
-        stdoutfile = open(stdoutname, "wb")
+        if "task.stdout" in taskp:
+            stdoutname = subst.do_substitution(taskp, taskp["task.stdout"])
+            stdoutfile = open(stdoutname, "wb")
 
-    logging.info("{}{}{}".format(' '.join(cmd), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else ""))
+    logger.info("{}{}{}".format(' '.join(cmd), (" < " + stdinname) if stdinname is not None else "", (" > " + stdoutname) if stdoutname is not None else ""))
 
+    if args.dry_run:
+        sys.exit(0)
+except subst.SubstitutionError as e:
+    logger.error(str(e))
+    logger.error("task parameters were:")
+    logger.error(pprint.pformat(taskp))
+    sys.exit(1)
 except Exception as e:
-    logging.exception("caught exception")
-    logging.error("task parameters was:")
-    logging.error(pprint.pformat(taskp))
+    logger.exception("caught exception")
+    logger.error("task parameters were:")
+    logger.error(pprint.pformat(taskp))
     sys.exit(1)
 
 try:
@@ -185,13 +274,13 @@ try:
     rcode = sp.wait()
 
     if sig.sig is not None:
-        logging.critical("terminating on signal %s" % sig.sig)
+        logger.critical("terminating on signal %s" % sig.sig)
         sys.exit(2)
     else:
-        logging.info("completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
+        logger.info("completed with exit code %i (%s)" % (rcode, "success" if rcode == 0 else "failed"))
 
 except Exception as e:
-    logging.exception("caught exception")
+    logger.exception("caught exception")
 
 # restore default signal handlers.
 signal.signal(signal.SIGINT, signal.SIG_DFL)
@@ -201,11 +290,11 @@ signal.signal(signal.SIGQUIT, signal.SIG_DFL)
 for l in links:
     os.unlink(l)
 
-logging.info("the following output files will be saved to keep:")
+logger.info("the following output files will be saved to keep:")
 
 subprocess.call(["find", ".", "-type", "f", "-printf", "run-command: %12.12s %h/%f\\n"], stdout=sys.stderr)
 
-logging.info("start writing output to keep")
+logger.info("start writing output to keep")
 
 if "task.vwd" in taskp:
     if "task.foreach" in jobp:
@@ -215,7 +304,7 @@ if "task.vwd" in taskp:
         # Just a single task, so do merge with the original collection
         outcollection = vwd.checkin(subst.do_substitution(taskp, taskp["task.vwd"]), outdir, merge=True).manifest_text()
 else:
-    outcollection = robust_put.upload(outdir)
+    outcollection = robust_put.upload(outdir, logger)
 
 api.job_tasks().update(uuid=arvados.current_task()['uuid'],
                                      body={