git://git.onelab.eu / monitor.git / blobdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
www/printbadnodes.py
[monitor.git] / bootman.py
diff --git a/bootman.py b/bootman.py
index b9a161f..a278afe 100755 (executable)
--- a/bootman.py
+++ b/bootman.py
@@ -7,7 +7,7 @@ api = plc.getAuthAPI()
 
 import sys
 import os
-import policy
+import const
 
 from getsshkeys import SSHKnownHosts
 
@@ -321,7 +321,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                        mailtxt.newbootcd_one[1] % args, True, db='bootcd_persistmessages')
 
                loginbase = plc.siteId(hostname)
-               m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+               m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
 
                print "\tDisabling %s due to out-of-date BOOTCD" % hostname
                api.UpdateNode(hostname, {'boot_state' : 'disable'})
@@ -400,25 +400,34 @@ def reboot(hostname, config=None, forced_action=None):
                        ('ccisserror' , 'cciss: cmd \w+ has CHECK CONDITION  byte \w+ = \w+'),
 
                        ('buffererror', 'Buffer I/O error on device dm-\d, logical block \d+'),
+
+                       ('hdaseekerror', 'hda: dma_intr: status=0x\d+ { DriveReady SeekComplete Error }'),
+                       ('hdacorrecterror', 'hda: dma_intr: error=0x\d+ { UncorrectableError }, LBAsect=\d+, sector=\d+'),
+
                        ('atareadyerror'   , 'ata\d+: status=0x\d+ { DriveReady SeekComplete Error }'),
                        ('atacorrecterror' , 'ata\d+: error=0x\d+ { UncorrectableError }'),
+
                        ('sdXerror'   , 'sd\w: Current: sense key: Medium Error'),
                        ('ext3error'   , 'EXT3-fs error (device dm-\d+): ext3_find_entry: reading directory #\d+ offset \d+'),
+
                        ('floppytimeout','floppy0: floppy timeout called'),
                        ('floppyerror',  'end_request: I/O error, dev fd\w+, sector \d+'),
 
+                       # hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
+                       # hda: dma_intr: error=0x40 { UncorrectableError }, LBAsect=23331263, sector=23331263
+
                        # floppy0: floppy timeout called
                        # end_request: I/O error, dev fd0, sector 0
 
-                       #Buffer I/O error on device dm-2, logical block 8888896
-                       #ata1: status=0x51 { DriveReady SeekComplete Error }
-                       #ata1: error=0x40 { UncorrectableError }
-                       #SCSI error : <0 0 0 0> return code = 0x8000002
-                       #sda: Current: sense key: Medium Error
+                       # Buffer I/O error on device dm-2, logical block 8888896
+                       # ata1: status=0x51 { DriveReady SeekComplete Error }
+                       # ata1: error=0x40 { UncorrectableError }
+                       # SCSI error : <0 0 0 0> return code = 0x8000002
+                       # sda: Current: sense key: Medium Error
                        #       Additional sense: Unrecovered read error - auto reallocate failed
 
-                       #SCSI error : <0 2 0 0> return code = 0x40001
-                       #end_request: I/O error, dev sda, sector 572489600
+                       # SCSI error : <0 2 0 0> return code = 0x40001
+                       # end_request: I/O error, dev sda, sector 572489600
                ]
                id = index_to_id(steps, child.expect( steps_to_list(steps) + [ pexpect.EOF ]))
                sequence.append(id)
@@ -444,7 +453,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddisk[1] % args, True, db='hardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.set_nodestate('disable')
                        return False
 
@@ -606,11 +615,14 @@ def reboot(hostname, config=None, forced_action=None):
 
        # update_node_config_email
        for n in ["bminit-cfg-exception-nocfg-update-bootupdatefail-nonode-debug-done",
-                       "bminit-cfg-exception-update-bootupdatefail-nonode-debug-done",
+                         "bminit-cfg-exception-update-bootupdatefail-nonode-debug-done",
+                         "bminit-cfg-auth-bootcheckfail-nonode-exception-update-bootupdatefail-nonode-debug-done",
                        ]:
                sequences.update({n : "update_node_config_email"})
 
-       for n in [ "bminit-cfg-exception-nodehostname-update-debug-done", ]:
+       for n in [ "bminit-cfg-exception-nodehostname-update-debug-done", 
+                          "bminit-cfg-update-exception-nodehostname-update-debug-done", 
+                       ]:
                sequences.update({n : "nodenetwork_email"})
 
        # update_bootcd_email
@@ -634,7 +646,11 @@ def reboot(hostname, config=None, forced_action=None):
        sequences.update({"bminit-cfg-auth-getplc-update-hardware-exception-hardwarerequirefail-update-debug-done" : "broken_hardware_email"})
 
        # bad_dns_email
-       sequences.update({"bminit-cfg-update-implementerror-bootupdatefail-dnserror-update-implementerror-bootupdatefail-dnserror-done" : "bad_dns_email"})
+       for n in [ 
+        "bminit-cfg-update-implementerror-bootupdatefail-dnserror-update-implementerror-bootupdatefail-dnserror-done",
+               "bminit-cfg-auth-implementerror-bootcheckfail-dnserror-update-implementerror-bootupdatefail-dnserror-done",
+               ]:
+               sequences.update( { n : "bad_dns_email"})
 
        flag_set = True
 
@@ -699,7 +715,7 @@ def reboot(hostname, config=None, forced_action=None):
                        m = PersistMessage(hostname,  mailtxt.plnode_cfg[0] % args,  mailtxt.plnode_cfg[1] % args, 
                                                                True, db='nodeid_persistmessages')
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.dump_plconf_file()
                        conn.set_nodestate('disable')
 
@@ -711,7 +727,7 @@ def reboot(hostname, config=None, forced_action=None):
                        m = PersistMessage(hostname,  mailtxt.plnode_network[0] % args,  mailtxt.plnode_cfg[1] % args, 
                                                                True, db='nodenet_persistmessages')
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.dump_plconf_file()
                        conn.set_nodestate('disable')
 
@@ -726,7 +742,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                                mailtxt.newalphacd_one[1] % args, True, db='bootcd_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
 
                        print "\tDisabling %s due to out-of-date BOOTCD" % hostname
                        conn.set_nodestate('disable')
@@ -744,7 +760,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddisk[1] % args, True, db='hardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.set_nodestate('disable')
 
                elif sequences[s] == "update_hardware_email":
@@ -756,7 +772,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.minimalhardware[1] % args, True, db='minhardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.set_nodestate('disable')
 
                elif sequences[s] == "bad_dns_email":
@@ -779,7 +795,7 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddns[1] % args, True, db='baddns_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([policy.PIEMAIL % loginbase, policy.TECHEMAIL % loginbase])
+                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
                        conn.set_nodestate('disable')
 
        if flag_set:
monitor