changes for 3.0

[monitor.git] / findbadpcu.py

diff --git a/findbadpcu.py b/findbadpcu.py
index e3d160d..114c48b 100755 (executable)
--- a/findbadpcu.py
+++ b/findbadpcu.py
@@ -12,6 +12,7 @@ import sets
     
 import signal
 import traceback
+from nodequery import pcu_select
 
 #old_handler = signal.getsignal(signal.SIGCHLD)
 
@@ -84,6 +85,8 @@ def get_pcu(pcuname):
                                        l_pcu = i
                except:
                        traceback.print_exc()
+                       from nodecommon import email_exception
+                       email_exception()
                        l_pcu = None
 
        plc_lock.release()
@@ -102,6 +105,8 @@ def get_nodes(node_ids):
                                        l_node.append(n)
                except:
                        traceback.print_exc()
+                       from nodecommon import email_exception
+                       email_exception()
                        l_node = None
 
        plc_lock.release()
@@ -159,6 +164,8 @@ def get_plc_site_values(site_id):
                                        break
                except:
                        traceback.print_exc()
+                       from nodecommon import email_exception
+                       email_exception()
                        values = None
 
        plc_lock.release()
@@ -197,6 +204,8 @@ def collectPingAndSSH(pcuname, cohash):
                except:
                        b_except = True
                        traceback.print_exc()
+                       from nodecommon import email_exception
+                       email_exception()
                        continue_probe = False
 
                if b_except or not continue_probe: return (None, None, None)
@@ -329,7 +338,7 @@ def checkAndRecordState(l_pcus, cohash):
        global count
        global_round = externalState['round']
 
-       tp = threadpool.ThreadPool(20)
+       tp = threadpool.ThreadPool(10)
 
        # CREATE all the work requests
        for pcuname in l_pcus:
@@ -390,6 +399,11 @@ def main():
                pcus = []
                for node in l_nodes:
                        pcus += node['pcu_ids']
+               # clear out dups.
+               l_pcus = [pcu for pcu in sets.Set(pcus)]
+       elif config.pcuselect is not None:
+               n, pcus = pcu_select(config.pcuselect)
+               # clear out dups.
                l_pcus = [pcu for pcu in sets.Set(pcus)]
 
        elif config.nodelist == None and config.pcuid == None:
@@ -421,6 +435,7 @@ if __name__ == '__main__':
        parser.set_defaults(nodelist=None, 
                                                increment=False, 
                                                pcuid=None,
+                                               pcuselect=None,
                                                site=None,
                                                dbname="findbadpcus", 
                                                cachenodes=False,
@@ -430,6 +445,8 @@ if __name__ == '__main__':
                                                help="Provide the input file for the node list")
        parser.add_option("", "--site", dest="site", metavar="FILE", 
                                                help="Get all pcus associated with the given site's nodes")
+       parser.add_option("", "--pcuselect", dest="pcuselect", metavar="FILE", 
+                                               help="Query string to apply to the findbad pcus")
        parser.add_option("", "--pcuid", dest="pcuid", metavar="id", 
                                                help="Provide the id for a single pcu")
 
@@ -452,6 +469,8 @@ if __name__ == '__main__':
                time.sleep(1)
        except Exception, err:
                traceback.print_exc()
+               from nodecommon import email_exception
+               email_exception()
                print "Exception: %s" % err
                print "Saving data... exitting."
                database.dbDump(config.dbname, externalState)