X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=policy.py;h=26187ddc4dbf32b2c4bc5dce19d4be5f538f75c2;hb=19414270cf2c8429daab02fdebbd8081d9ba0db0;hp=e99fb71284b9981eca5c21f0505ff4a15f6094b9;hpb=208bab758d8ade2c3948f88c297df0120af2fa42;p=monitor.git diff --git a/policy.py b/policy.py index e99fb71..26187dd 100644 --- a/policy.py +++ b/policy.py @@ -3,7 +3,7 @@ # # Faiyaz Ahmed # -# $Id: policy.py,v 1.15 2007/07/03 19:58:34 soltesz Exp $ +# $Id: policy.py,v 1.17 2007/08/29 17:26:50 soltesz Exp $ # # Policy Engine. @@ -17,13 +17,12 @@ import pickle import Queue import plc import sys +import os import reboot -import soltesz +import database import string -from printbadbysite import cmpCategoryVal -from config import config -print "policy" -config = config() +from unified_model import cmpCategoryVal +import config DAT="./monitor.dat" @@ -40,6 +39,8 @@ SLICEMAIL="%s@slices.planet-lab.org" PLCEMAIL="support@planet-lab.org" #Thresholds (DAYS) +SPERMIN = 60 +SPERHOUR = 60*60 SPERDAY = 86400 PITHRESH = 7 * SPERDAY SLICETHRESH = 7 * SPERDAY @@ -81,18 +82,22 @@ def array_to_priority_map(array): def getdebug(): return config.debug +def print_stats(key, stats): + if key in stats: print "%20s : %d" % (key, stats[key]) + + class Merge(Thread): def __init__(self, l_merge, toRT): self.toRT = toRT self.merge_list = l_merge # the hostname to loginbase mapping - self.plcdb_hn2lb = soltesz.dbLoad("plcdb_hn2lb") + self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb") # Previous actions taken on nodes. - self.act_all = soltesz.if_cached_else(1, "act_all", lambda : {}) - self.findbad = soltesz.if_cached_else(1, "findbad", lambda : {}) + self.act_all = database.if_cached_else(1, "act_all", lambda : {}) + self.findbad = database.if_cached_else(1, "findbad", lambda : {}) - self.cache_all = soltesz.if_cached_else(1, "act_all", lambda : {}) + self.cache_all = database.if_cached_else(1, "act_all", lambda : {}) self.sickdb = {} self.mergedb = {} Thread.__init__(self) @@ -125,9 +130,17 @@ class Merge(Thread): fb_record = {} fb_record['nodename'] = nodename - fb_record['category'] = values['category'] + try: + fb_record['category'] = values['category'] + except: + print values + print nodename + print self.findbad['nodes'][nodename] + count -= 1 + continue fb_record['state'] = values['state'] fb_record['comonstats'] = values['comonstats'] + fb_record['plcnode'] = values['plcnode'] fb_record['kernel'] = self.getKernel(values['kernel']) fb_record['stage'] = "findbad" fb_record['message'] = None @@ -185,64 +198,44 @@ class Merge(Thread): if loginbase not in self.mergedb: self.mergedb[loginbase] = {} + # take the info either from act_all or fb-record. + # if node not in act_all + # then take it from fbrecord, obviously. + # else node in act_all + # if act_all == 0 length (no previous records) + # then take it from fbrecord. + # else + # take it from act_all. + # + # We must compare findbad state with act_all state if nodename not in self.act_all: # 1) ok, b/c it's a new problem. set ticket_id to null self.mergedb[loginbase][nodename] = {} self.mergedb[loginbase][nodename].update(x) self.mergedb[loginbase][nodename]['ticket_id'] = "" - self.mergedb[loginbase][nodename]['prev_category'] = None + self.mergedb[loginbase][nodename]['prev_category'] = "NORECORD" else: - y = self.act_all[nodename][0] - - # skip if end-stage - if 'stage' in y and "monitor-end-record" in y['stage']: - continue - - # for legacy actions - if 'bucket' in y and y['bucket'][0] == 'dbg': - # Only bootcd debugs made it to the act_all db. - y['prev_category'] = "OLDBOOTCD" - elif 'bucket' in y and y['bucket'][0] == 'down': - y['prev_category'] = "ERROR" - elif 'bucket' not in y: - # for all other actions, just carry over the - # previous category - y['prev_category'] = y['category'] + if len(self.act_all[nodename]) == 0: + self.mergedb[loginbase][nodename] = {} + self.mergedb[loginbase][nodename].update(x) + self.mergedb[loginbase][nodename]['ticket_id'] = "" + self.mergedb[loginbase][nodename]['prev_category'] = "NORECORD" else: - print "UNKNOWN state for record: %s" % y - sys.exit(1) - # determine through translation, if the buckets match - #if 'category' in y and x['category'] == y['category']: - # b_match = True - #elif x['category'] == "OLDBOOTCD" and y['bucket'][0] == 'dbg': - # b_match = True - #elif x['category'] == "ERROR" and y['bucket'][0] == 'down': - # b_match = True - #else: - # b_match = False - - #if b_match: - # # 2b) ok, b/c they agree that there's still a problem.. - # # 2b) Comon & Monitor still agree; RT ticket? - # y['prev_category'] = y['category'] - #else: - # # 2a) mismatch, need a policy for how to resolve - # # resolution will be handled in __diagnoseNode() - # # for now just record the two categories. - # #if x['category'] == "PROD" and x['state'] == "BOOT" and \ - # # ( y['bucket'][0] == 'down' or y['bucket'][0] == 'dbg'): - # print "FINDBAD and MONITOR have a mismatch: %s vs %s" % \ - # (x['category'], y['bucket']) - - - self.mergedb[loginbase][nodename] = {} - self.mergedb[loginbase][nodename].update(y) - self.mergedb[loginbase][nodename]['comonstats'] = x['comonstats'] - self.mergedb[loginbase][nodename]['category'] = x['category'] - self.mergedb[loginbase][nodename]['state'] = x['state'] - self.mergedb[loginbase][nodename]['kernel']=x['kernel'] - self.mergedb[loginbase][nodename]['bootcd']=x['bootcd'] + y = self.act_all[nodename][0] + y['prev_category'] = y['category'] + + self.mergedb[loginbase][nodename] = {} + self.mergedb[loginbase][nodename].update(y) + self.mergedb[loginbase][nodename]['comonstats'] = x['comonstats'] + self.mergedb[loginbase][nodename]['category'] = x['category'] + self.mergedb[loginbase][nodename]['state'] = x['state'] + self.mergedb[loginbase][nodename]['kernel']=x['kernel'] + self.mergedb[loginbase][nodename]['bootcd']=x['bootcd'] + self.mergedb[loginbase][nodename]['plcnode']=x['plcnode'] + ticket = get_ticket_id(self.mergedb[loginbase][nodename]) + self.mergedb[loginbase][nodename]['rt'] = mailer.getTicketStatus(ticket) + # delete the entry from cache_all to keep it out of case 3) del self.cache_all[nodename] @@ -282,14 +275,13 @@ class Merge(Thread): class Diagnose(Thread): def __init__(self, fromRT): self.fromRT = fromRT - self.plcdb_hn2lb = soltesz.dbLoad("plcdb_hn2lb") + self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb") + self.findbad = database.if_cached_else(1, "findbad", lambda : {}) self.diagnose_in = {} self.diagnose_out = {} Thread.__init__(self) - def print_stats(self, key, stats): - print "%20s : %d" % (key, stats[key]) def run(self): self.accumSickSites() @@ -307,13 +299,13 @@ class Diagnose(Thread): #if config.policysavedb: sys.exit(1) - self.print_stats("sites", stats) - self.print_stats("sites_diagnosed", stats) - self.print_stats("nodes_diagnosed", stats) + print_stats("sites_observed", stats) + print_stats("sites_diagnosed", stats) + print_stats("nodes_diagnosed", stats) if config.policysavedb: print "Saving Databases... diagnose_out" - soltesz.dbDump("diagnose_out", self.diagnose_out) + database.dbDump("diagnose_out", self.diagnose_out) def accumSickSites(self): """ @@ -338,7 +330,7 @@ class Diagnose(Thread): return def diagnoseAll(self): - i_sites = 0 + i_sites_observed = 0 i_sites_diagnosed = 0 i_nodes_diagnosed = 0 i_nodes_actedon = 0 @@ -347,52 +339,82 @@ class Diagnose(Thread): sorted_sites = self.diagnose_in.keys() sorted_sites.sort() - l_diagnosed_all = [] + self.diagnose_out= {} for loginbase in sorted_sites: l_allsites += [loginbase] d_diag_nodes = self.diagnose_in[loginbase] - l_diag_records = self.__diagnoseSite(loginbase, d_diag_nodes) - l_diagnosed_all += l_diag_records + d_act_records = self.__diagnoseSite(loginbase, d_diag_nodes) + # store records in diagnose_out, for saving later. + self.diagnose_out.update(d_act_records) - if len(l_diag_records) > 0: - i_nodes_diagnosed += len(l_diag_records) + if len(d_act_records[loginbase]['nodes'].keys()) > 0: + i_nodes_diagnosed += (len(d_act_records[loginbase]['nodes'].keys())) i_sites_diagnosed += 1 - i_sites += 1 - - self.diagnose_out= {} - for diag_record in l_diagnosed_all: - nodename = diag_record['nodename'] - loginbase = self.plcdb_hn2lb[nodename] + i_sites_observed += 1 - if loginbase not in self.diagnose_out: - self.diagnose_out[loginbase] = {} - - self.diagnose_out[loginbase][nodename] = diag_record - - return {'sites': i_sites, + return {'sites_observed': i_sites_observed, 'sites_diagnosed': i_sites_diagnosed, 'nodes_diagnosed': i_nodes_diagnosed, 'allsites':l_allsites} pass - def __getDaysDown(self, diag_record, nodename): + def getDaysDown(cls, diag_record): daysdown = -1 - if diag_record['comonstats']['sshstatus'] != "null": - daysdown = int(diag_record['comonstats']['sshstatus']) // (60*60*24) - elif diag_record['comonstats']['lastcotop'] != "null": - daysdown = int(diag_record['comonstats']['lastcotop']) // (60*60*24) + last_contact = diag_record['plcnode']['last_contact'] + date_created = diag_record['plcnode']['date_created'] + + if diag_record['comonstats']['uptime'] != "null" and diag_record['comonstats']['uptime'] != "-1": + daysdown = - int(float(diag_record['comonstats']['uptime'])) // (60*60*24) + elif last_contact is None: + if date_created is not None: + now = time.time() + diff = now - date_created + daysdown = diff // (60*60*24) + else: + daysdown = -1 else: - daysdown = -1 + now = time.time() + diff = now - last_contact + daysdown = diff // (60*60*24) return daysdown - - def __getStrDaysDown(self, diag_record, nodename): - daysdown = self.__getDaysDown(diag_record, nodename) - if daysdown > 0: - return "(%d days down)"%daysdown + getDaysDown = classmethod(getDaysDown) + + def getStrDaysDown(cls, diag_record): + daysdown = "unknown" + last_contact = diag_record['plcnode']['last_contact'] + date_created = diag_record['plcnode']['date_created'] + + if diag_record['comonstats']['uptime'] != "null" and \ + diag_record['comonstats']['uptime'] != "-1": + daysdown = int(float(diag_record['comonstats']['uptime'])) // (60*60*24) + daysdown = "%d days up" % daysdown + + elif last_contact is None: + if date_created is not None: + now = time.time() + diff = now - date_created + daysdown = diff // (60*60*24) + daysdown = "Never contacted PLC, created %s days ago" % daysdown + else: + daysdown = "Never contacted PLC" else: - return "Unknown number of days" + now = time.time() + diff = now - last_contact + daysdown = diff // (60*60*24) + daysdown = "%s days down" % daysdown + return daysdown + getStrDaysDown = classmethod(getStrDaysDown) + #def getStrDaysDown(cls, diag_record): + # daysdown = cls.getDaysDown(diag_record) + # if daysdown > -1: + # return "%d days down"%daysdown + # elif daysdown == -1: + # return "Has never contacted PLC" + # else: + # return "%d days up"% -daysdown + #getStrDaysDown = classmethod(getStrDaysDown) def __getCDVersion(self, diag_record, nodename): cdversion = "" @@ -402,9 +424,15 @@ class Diagnose(Thread): def __diagnoseSite(self, loginbase, d_diag_nodes): """ - rec_sitelist is a diagnose_in entry: + d_diag_nodes are diagnose_in entries. """ - diag_list = [] + d_diag_site = {loginbase : { 'config' : + {'squeeze': False, + 'email': False + }, + 'nodes': {} + } + } sorted_nodes = d_diag_nodes.keys() sorted_nodes.sort() for nodename in sorted_nodes: @@ -412,9 +440,37 @@ class Diagnose(Thread): diag_record = self.__diagnoseNode(loginbase, node_record) if diag_record != None: - diag_list += [ diag_record ] + d_diag_site[loginbase]['nodes'][nodename] = diag_record + + # NOTE: improvement means, we need to act/squeeze and email. + #print "DIAG_RECORD", diag_record + if 'monitor-end-record' in diag_record['stage'] or \ + 'nmreset' in diag_record['stage']: + # print "resetting loginbase!" + d_diag_site[loginbase]['config']['squeeze'] = True + d_diag_site[loginbase]['config']['email'] = True + #else: + # print "NO IMPROVEMENT!!!!" + else: + pass # there is nothing to do for this node. - return diag_list + # NOTE: these settings can be overridden by command line arguments, + # or the state of a record, i.e. if already in RT's Support Queue. + nodes_up = self.getUpAtSite(loginbase, d_diag_site) + if nodes_up < MINUP: + d_diag_site[loginbase]['config']['squeeze'] = True + + max_slices = self.getMaxSlices(loginbase) + num_nodes = self.getNumNodes(loginbase) + # NOTE: when max_slices == 0, this is either a new site (the old way) + # or an old disabled site from previous monitor (before site['enabled']) + if nodes_up < num_nodes and max_slices != 0: + d_diag_site[loginbase]['config']['email'] = True + + if len(d_diag_site[loginbase]['nodes'].keys()) > 0: + print "SITE: %20s : %d nodes up, at most" % (loginbase, nodes_up) + + return d_diag_site def diagRecordByCategory(self, node_record): nodename = node_record['nodename'] @@ -426,22 +482,44 @@ class Diagnose(Thread): if "ERROR" in category: # i.e. "DOWN" diag_record = {} diag_record.update(node_record) - daysdown = self.__getDaysDown(diag_record, nodename) - if daysdown >= 0 and daysdown < 7: + daysdown = self.getDaysDown(diag_record) + if daysdown < 7: format = "DIAG: %20s : %-40s Down only %s days NOTHING DONE" print format % (loginbase, nodename, daysdown) return None - s_daysdown = self.__getStrDaysDown(diag_record, nodename) + s_daysdown = self.getStrDaysDown(diag_record) diag_record['message'] = emailTxt.mailtxt.newdown diag_record['args'] = {'nodename': nodename} diag_record['info'] = (nodename, s_daysdown, "") + + if 'reboot_node_failed' in node_record: + # there was a previous attempt to use the PCU. + if node_record['reboot_node_failed'] == False: + # then the last attempt apparently, succeeded. + # But, the category is still 'ERROR'. Therefore, the + # PCU-to-Node mapping is broken. + #print "Setting message for ERROR node to PCU2NodeMapping: %s" % nodename + diag_record['message'] = emailTxt.mailtxt.pcutonodemapping + diag_record['email_pcu'] = True + + if 'ticket_id' in diag_record: + if diag_record['ticket_id'] == "": + if 'found_rt_ticket' in diag_record: + ticket_id = diag_record['found_rt_ticket'] + else: + ticket_id = "None" + else: + ticket_id = diag_record['ticket_id'] + else: + ticket_id = "None" + diag_record['log'] = "DOWN: %20s : %-40s == %20s %s" % \ - (loginbase, nodename, diag_record['info'], diag_record['ticket_id']), + (loginbase, nodename, diag_record['info'][1:], ticket_id) elif "OLDBOOTCD" in category: # V2 boot cds as determined by findbad - s_daysdown = self.__getStrDaysDown(node_record, nodename) + s_daysdown = self.getStrDaysDown(node_record) s_cdversion = self.__getCDVersion(node_record, nodename) diag_record = {} diag_record.update(node_record) @@ -449,9 +527,14 @@ class Diagnose(Thread): diag_record['message'] = emailTxt.mailtxt.newbootcd diag_record['args'] = {'nodename': nodename} diag_record['info'] = (nodename, s_daysdown, s_cdversion) - diag_record['log'] = "BTCD: %20s : %-40s == %20s %20s %s" % \ + if diag_record['ticket_id'] == "": + diag_record['log'] = "BTCD: %20s : %-40s == %20s %20s %s" % \ (loginbase, nodename, diag_record['kernel'], - diag_record['bootcd'], diag_record['ticket_id']), + diag_record['bootcd'], diag_record['found_rt_ticket']) + else: + diag_record['log'] = "BTCD: %20s : %-40s == %20s %20s %s" % \ + (loginbase, nodename, diag_record['kernel'], + diag_record['bootcd'], diag_record['ticket_id']) elif "PROD" in category: if "DEBUG" in state: @@ -462,6 +545,13 @@ class Diagnose(Thread): elif "BOOT" in state: # no action needed. # TODO: remove penalties, if any are applied. + now = time.time() + last_contact = node_record['plcnode']['last_contact'] + if last_contact == None: + time_diff = 0 + else: + time_diff = now - last_contact; + if 'improvement' in node_record['stage']: # then we need to pass this on to 'action' diag_record = {} @@ -470,10 +560,44 @@ class Diagnose(Thread): diag_record['args'] = {'nodename': nodename} diag_record['info'] = (nodename, node_record['prev_category'], node_record['category']) - diag_record['log'] = "IMPR: %20s : %-40s == %20s %20s %s" % \ + if 'email_pcu' in diag_record: + if diag_record['email_pcu']: + # previously, the pcu failed to reboot, so send + # email. Now, reset these values to try the reboot + # again. + diag_record['email_pcu'] = False + del diag_record['reboot_node_failed'] + + if diag_record['ticket_id'] == "": + diag_record['log'] = "IMPR: %20s : %-40s == %20s %20s %s %s" % \ (loginbase, nodename, diag_record['stage'], - state, category), + state, category, diag_record['found_rt_ticket']) + else: + diag_record['log'] = "IMPR: %20s : %-40s == %20s %20s %s %s" % \ + (loginbase, nodename, diag_record['stage'], + state, category, diag_record['ticket_id']) return diag_record + #elif time_diff >= 6*SPERHOUR: + # # heartbeat is older than 30 min. + # # then reset NM. + # #print "Possible NM problem!! %s - %s = %s" % (now, last_contact, time_diff) + # diag_record = {} + # diag_record.update(node_record) + # diag_record['message'] = emailTxt.mailtxt.NMReset + # diag_record['args'] = {'nodename': nodename} + # diag_record['stage'] = "nmreset" + # diag_record['info'] = (nodename, + # node_record['prev_category'], + # node_record['category']) + # if diag_record['ticket_id'] == "": + # diag_record['log'] = "NM : %20s : %-40s == %20s %20s %s %s" % \ + # (loginbase, nodename, diag_record['stage'], + # state, category, diag_record['found_rt_ticket']) + # else: + # diag_record['log'] = "NM : %20s : %-40s == %20s" % \ + # (loginbase, nodename, diag_record['stage']) +# +# return diag_record else: return None else: @@ -496,22 +620,61 @@ class Diagnose(Thread): def __diagnoseNode(self, loginbase, node_record): # TODO: change the format of the hostname in this # record to something more natural. - nodename = node_record['nodename'] - category = node_record['category'] - prev_category = node_record['prev_category'] - state = node_record['state'] - - val = cmpCategoryVal(category, prev_category) - if val == -1: - # current category is worse than previous, carry on - pass - elif val == 1: - # current category is better than previous - # TODO: too generous for now, but will be handled correctly - node_record['stage'] = 'improvement' - else: - #values are equal, carry on. - pass + nodename = node_record['nodename'] + category = node_record['category'] + prev_category = node_record['prev_category'] + state = node_record['state'] + #if 'prev_category' in node_record: + # prev_category = node_record['prev_category'] + #else: + # prev_category = "ERROR" + if node_record['prev_category'] != "NORECORD": + + val = cmpCategoryVal(category, prev_category) + print "%s went from %s -> %s" % (nodename, prev_category, category) + if val == 1: + # improved + if node_record['ticket_id'] == "" or node_record['ticket_id'] == None: + print "closing record with no ticket: ", node_record['nodename'] + node_record['action'] = ['close_rt'] + node_record['message'] = None + node_record['stage'] = 'monitor-end-record' + return node_record + else: + node_record['stage'] = 'improvement' + + #if 'monitor-end-record' in node_record['stage']: + # # just ignore it if it's already ended. + # # otherwise, the status should be worse, and we won't get + # # here. + # print "monitor-end-record: ignoring ", node_record['nodename'] + # return None +# +# #return None + elif val == -1: + # current category is worse than previous, carry on + pass + else: + #values are equal, carry on. + #print "why are we here?" + pass + + if 'rt' in node_record and 'Status' in node_record['rt']: + if node_record['stage'] == 'ticket_waitforever': + if 'resolved' in node_record['rt']['Status']: + print "ending waitforever record for: ", node_record['nodename'] + node_record['action'] = ['noop'] + node_record['message'] = None + node_record['stage'] = 'monitor-end-record' + print "oldlog: %s" % node_record['log'], + print "%15s" % node_record['action'] + return node_record + if 'new' in node_record['rt']['Status'] and \ + 'Queue' in node_record['rt'] and \ + 'Monitor' in node_record['rt']['Queue']: + + print "RESETTING stage to findbad" + node_record['stage'] = 'findbad' #### COMPARE category and prev_category # if not_equal @@ -520,16 +683,24 @@ class Diagnose(Thread): # then check category for stats. diag_record = self.diagRecordByCategory(node_record) if diag_record == None: + #print "diag_record == None" return None #### found_RT_ticket # TODO: need to record time found, and maybe add a stage for acting on it... + # NOTE: after found, if the support ticket is resolved, the block is + # not removed. How to remove the block on this? if 'found_rt_ticket' in diag_record and \ diag_record['found_rt_ticket'] is not None: if diag_record['stage'] is not 'improvement': diag_record['stage'] = 'ticket_waitforever' current_time = time.time() + # take off four days, for the delay that database caused. + # TODO: generalize delays at PLC, and prevent enforcement when there + # have been no emails. + # NOTE: 7*SPERDAY exists to offset the 'bad week' + #delta = current_time - diag_record['time'] - 7*SPERDAY delta = current_time - diag_record['time'] message = diag_record['message'] @@ -537,77 +708,104 @@ class Diagnose(Thread): act_record.update(diag_record) #### DIAGNOSE STAGES - #print "%s has stage %s" % (nodename, diag_record['stage']) if 'findbad' in diag_record['stage']: # The node is bad, and there's no previous record of it. - act_record['email'] = TECH # addative emails - act_record['action'] = 'noop' + act_record['email'] = TECH + act_record['action'] = ['noop'] act_record['message'] = message[0] act_record['stage'] = 'stage_actinoneweek' + elif 'nmreset' in diag_record['stage']: + act_record['email'] = ADMIN + act_record['action'] = ['reset_nodemanager'] + act_record['message'] = message[0] + act_record['stage'] = 'nmreset' + return None + + elif 'reboot_node' in diag_record['stage']: + act_record['email'] = TECH + act_record['action'] = ['noop'] + act_record['message'] = message[0] + act_record['stage'] = 'stage_actinoneweek' + elif 'improvement' in diag_record['stage']: # - backoff previous squeeze actions (slice suspend, nocreate) # TODO: add a backoff_squeeze section... Needs to runthrough - act_record['action'] = 'close_rt' + print "backing off of %s" % nodename + act_record['action'] = ['close_rt'] act_record['message'] = message[0] act_record['stage'] = 'monitor-end-record' elif 'actinoneweek' in diag_record['stage']: - act_record['email'] = TECH | PI # addative emails if delta >= 7 * SPERDAY: + act_record['email'] = TECH | PI act_record['stage'] = 'stage_actintwoweeks' act_record['message'] = message[1] - act_record['action'] = 'nocreate' + act_record['action'] = ['nocreate' ] + act_record['time'] = current_time # reset clock for waitforever elif delta >= 3* SPERDAY and not 'second-mail-at-oneweek' in act_record: - act_record['message'] = message[1] - act_record['action'] = 'sendmailagain-waitforoneweekaction' + act_record['email'] = TECH + act_record['message'] = message[0] + act_record['action'] = ['sendmailagain-waitforoneweekaction' ] act_record['second-mail-at-oneweek'] = True else: act_record['message'] = None - act_record['action'] = 'waitforoneweekaction' + act_record['action'] = ['waitforoneweekaction' ] + print "ignoring this record for: %s" % act_record['nodename'] + return None # don't send if there's no action elif 'actintwoweeks' in diag_record['stage']: - act_record['email'] = TECH | PI | USER # addative emails - if delta >= 14 * SPERDAY: + if delta >= 7 * SPERDAY: + act_record['email'] = TECH | PI | USER act_record['stage'] = 'stage_waitforever' act_record['message'] = message[2] - act_record['action'] = 'suspendslices' + act_record['action'] = ['suspendslices'] act_record['time'] = current_time # reset clock for waitforever - elif delta >= 10* SPERDAY and not 'second-mail-at-twoweeks' in act_record: - act_record['message'] = message[2] - act_record['action'] = 'sendmailagain-waitfortwoweeksaction' + elif delta >= 3* SPERDAY and not 'second-mail-at-twoweeks' in act_record: + act_record['email'] = TECH | PI + act_record['message'] = message[1] + act_record['action'] = ['sendmailagain-waitfortwoweeksaction' ] act_record['second-mail-at-twoweeks'] = True else: act_record['message'] = None - act_record['action'] = 'waitfortwoweeksaction' + act_record['action'] = ['waitfortwoweeksaction'] + return None # don't send if there's no action elif 'ticket_waitforever' in diag_record['stage']: act_record['email'] = TECH if 'first-found' not in act_record: act_record['first-found'] = True - act_record['action'] = 'ticket_waitforever' - act_record['message'] = None + act_record['log'] += " firstfound" + act_record['action'] = ['ticket_waitforever'] + act_record['message'] = message[0] act_record['time'] = current_time else: if delta >= 7*SPERDAY: - act_record['action'] = 'email-againticket_waitforever' - act_record['message'] = message[0] + act_record['action'] = ['ticket_waitforever'] + if 'rt' in act_record and 'Status' in act_record['rt'] and \ + act_record['rt']['Status'] == 'new': + act_record['message'] = message[0] + else: + act_record['message'] = None + act_record['time'] = current_time # reset clock else: - act_record['action'] = 'ticket_waitforever' + act_record['action'] = ['ticket_waitforever'] act_record['message'] = None + return None elif 'waitforever' in diag_record['stage']: # more than 3 days since last action # TODO: send only on weekdays. # NOTE: expects that 'time' has been reset before entering waitforever stage if delta >= 3*SPERDAY: - act_record['action'] = 'email-againwaitforever' - act_record['message'] = message[0] + act_record['action'] = ['email-againwaitforever'] + act_record['message'] = message[2] act_record['time'] = current_time # reset clock else: - act_record['action'] = 'waitforever' + act_record['action'] = ['waitforever'] act_record['message'] = None + return None # don't send if there's no action else: # There is no action to be taken, possibly b/c the stage has @@ -616,16 +814,79 @@ class Diagnose(Thread): # 1. stage is unknown, or # 2. delta is not big enough to bump it to the next stage. # TODO: figure out which. for now assume 2. - print "UNKNOWN!!? %s" % nodename - act_record['action'] = 'unknown' + print "UNKNOWN stage for %s; nothing done" % nodename + act_record['action'] = ['unknown'] act_record['message'] = message[0] - print "Exiting..." - sys.exit(1) + + act_record['email'] = TECH + act_record['action'] = ['noop'] + act_record['message'] = message[0] + act_record['stage'] = 'stage_actinoneweek' + act_record['time'] = current_time # reset clock + #print "Exiting..." + #return None + #sys.exit(1) print "%s" % act_record['log'], print "%15s" % act_record['action'] return act_record + def getMaxSlices(self, loginbase): + # if sickdb has a loginbase, then it will have at least one node. + site_stats = None + + for nodename in self.diagnose_in[loginbase].keys(): + if nodename in self.findbad['nodes']: + site_stats = self.findbad['nodes'][nodename]['values']['plcsite'] + break + + if site_stats == None: + raise Exception, "loginbase with no nodes in findbad" + else: + return site_stats['max_slices'] + + def getNumNodes(self, loginbase): + # if sickdb has a loginbase, then it will have at least one node. + site_stats = None + + for nodename in self.diagnose_in[loginbase].keys(): + if nodename in self.findbad['nodes']: + site_stats = self.findbad['nodes'][nodename]['values']['plcsite'] + break + + if site_stats == None: + raise Exception, "loginbase with no nodes in findbad" + else: + if 'num_nodes' in site_stats: + return site_stats['num_nodes'] + else: + return 0 + + """ + Returns number of up nodes as the total number *NOT* in act_all with a + stage other than 'steady-state' . + """ + def getUpAtSite(self, loginbase, d_diag_site): + # TODO: THIS DOESN"T WORK!!! it misses all the 'debug' state nodes + # that aren't recorded yet. + + numnodes = self.getNumNodes(loginbase) + # NOTE: assume nodes we have no record of are ok. (too conservative) + # TODO: make the 'up' value more representative + up = numnodes + for nodename in d_diag_site[loginbase]['nodes'].keys(): + + rec = d_diag_site[loginbase]['nodes'][nodename] + if rec['stage'] != 'monitor-end-record': + up -= 1 + else: + pass # the node is assumed to be up. + + #if up != numnodes: + # print "ERROR: %s total nodes up and down != %d" % (loginbase, numnodes) + + return up + class SiteAction: def __init__(self, parameter_names=['hostname', 'ticket_id']): @@ -657,10 +918,10 @@ class BackoffActions(SiteAction): # TODO: create class for each action below, # allow for lists of actions to be performed... -def close_rt_backoff(args): - mailer.closeTicketViaRT(args['ticket_id'], "Ticket CLOSED automatically by SiteAssist.") - plc.enableSlices(args['hostname']) - plc.enableSliceCreation(args['hostname']) + + +def reset_nodemanager(args): + os.system("ssh root@%s /sbin/service nm restart" % nodename) return class Action(Thread): @@ -668,12 +929,12 @@ class Action(Thread): self.l_action = l_action # the hostname to loginbase mapping - self.plcdb_hn2lb = soltesz.dbLoad("plcdb_hn2lb") + self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb") # Actions to take. - self.diagnose_db = soltesz.if_cached_else(1, "diagnose_out", lambda : {}) + self.diagnose_db = database.if_cached_else(1, "diagnose_out", lambda : {}) # Actions taken. - self.act_all = soltesz.if_cached_else(1, "act_all", lambda : {}) + self.act_all = database.if_cached_else(1, "act_all", lambda : {}) # A dict of actions to specific functions. PICKLE doesnt' like lambdas. self.actions = {} @@ -682,6 +943,9 @@ class Action(Thread): self.actions['close_rt'] = lambda args: close_rt_backoff(args) self.actions['rins'] = lambda args: plc.nodeBootState(args['hostname'], "rins") self.actions['noop'] = lambda args: args + self.actions['reboot_node'] = lambda args: reboot_node(args) + self.actions['reset_nodemanager'] = lambda args: args # reset_nodemanager(args) + self.actions['ticket_waitforever'] = lambda args: args self.actions['waitforever'] = lambda args: args self.actions['unknown'] = lambda args: args @@ -710,22 +974,22 @@ class Action(Thread): print err if config.policysavedb: print "Saving Databases... act_all" - soltesz.dbDump("act_all", self.act_all) + database.dbDump("act_all", self.act_all) sys.exit(1) - self.print_stats("sites", stats) - self.print_stats("sites_diagnosed", stats) - self.print_stats("nodes_diagnosed", stats) - self.print_stats("sites_emailed", stats) - self.print_stats("nodes_actedon", stats) + print_stats("sites_observed", stats) + print_stats("sites_diagnosed", stats) + print_stats("nodes_diagnosed", stats) + print_stats("sites_emailed", stats) + print_stats("nodes_actedon", stats) print string.join(stats['allsites'], ",") if config.policysavedb: print "Saving Databases... act_all" - #soltesz.dbDump("policy.eventlog", self.eventlog) + #database.dbDump("policy.eventlog", self.eventlog) # TODO: remove 'diagnose_out', # or at least the entries that were acted on. - soltesz.dbDump("act_all", self.act_all) + database.dbDump("act_all", self.act_all) def accumSites(self): """ @@ -743,14 +1007,20 @@ class Action(Thread): loginbase = self.plcdb_hn2lb[nodename] if loginbase in self.diagnose_db and \ - nodename in self.diagnose_db[loginbase]: + nodename in self.diagnose_db[loginbase]['nodes']: - diag_record = self.diagnose_db[loginbase][nodename] + diag_record = self.diagnose_db[loginbase]['nodes'][nodename] if loginbase not in self.sickdb: - self.sickdb[loginbase] = {} - - self.sickdb[loginbase][nodename] = diag_record + self.sickdb[loginbase] = {'nodes' : {}} + + # NOTE: don't copy all node records, since not all will be in l_action + self.sickdb[loginbase]['nodes'][nodename] = diag_record + # NOTE: but, we want to get the loginbase config settings, + # this is the easiest way. + self.sickdb[loginbase]['config'] = self.diagnose_db[loginbase]['config'] + #else: + #print "%s not in diagnose_db!!" % loginbase return def __emailSite(self, loginbase, roles, message, args): @@ -796,14 +1066,6 @@ class Action(Thread): mailer.email(subj, body, contacts) ticket_id = args['ticket_id'] else: - #if 'ticket_id' in args and 'ticket_id' != "": - # # Reformat Subject to include Ticket_ID for RT - # subj = "Re: [PL #%s] %s" % (args['ticket_id'], subject) - # # RT remembers old contacts, so only add new users - # mailer.email(subj, body, ['monitor@planet-lab.org'] + contacts) - # ticket_id = args['ticket_id'] - #else: - # ticket_id = mailer.emailViaRT(subject, body, contacts) ticket_id = mailer.emailViaRT(subject, body, contacts, args['ticket_id']) except Exception, err: print "exception on message:" @@ -822,84 +1084,206 @@ class Action(Thread): hlist = " %s %s - %s\n" % (info[0], info[2], info[1]) #(node,ver,daysdn) return hlist - def __actOnSite(self, loginbase, site_record): - i_nodes_actedon = 0 - i_nodes_emailed = 0 - b_squeeze = config.squeeze - - act_recordlist = [] - for nodename in site_record.keys(): - diag_record = site_record[nodename] - act_record = self.__actOnNode(diag_record) - act_recordlist += [act_record] - - count_up = self.currentUpAtSite(loginbase) - if count_up < MINUP: - print "SITE: %20s : %d nodes up" % (loginbase, count_up) - else: - print "SITE: %20s : %d nodes up" % (loginbase, count_up) - # There may be a second penalty regardless of which stage it's in. - # TODO: check how long this has occurred. + def get_email_args(self, act_recordlist, loginbase=None): email_args = {} email_args['hostname_list'] = "" + for act_record in act_recordlist: email_args['hostname_list'] += act_record['msg_format'] email_args['hostname'] = act_record['nodename'] + if 'plcnode' in act_record and \ + 'pcu_ids' in act_record['plcnode'] and \ + len(act_record['plcnode']['pcu_ids']) > 0: + print "setting 'pcu_id' for email_args %s"%email_args['hostname'] + email_args['pcu_id'] = act_record['plcnode']['pcu_ids'][0] + else: + email_args['pcu_id'] = "-1" + if 'ticket_id' in act_record: - email_args['ticket_id'] = act_record['ticket_id'] + if act_record['ticket_id'] == 0 or act_record['ticket_id'] == '0': + print "Enter the ticket_id for %s @ %s" % (loginbase, act_record['nodename']) + sys.stdout.flush() + line = sys.stdin.readline() + try: + ticket_id = int(line) + except: + print "could not get ticket_id from stdin..." + os._exit(1) + else: + ticket_id = act_record['ticket_id'] + + email_args['ticket_id'] = ticket_id + + return email_args + + def get_unique_issues(self, act_recordlist): + # NOTE: only send one email per site, per problem... + unique_issues = {} + for act_record in act_recordlist: + act_key = act_record['action'][0] + if act_key not in unique_issues: + unique_issues[act_key] = [] + + unique_issues[act_key] += [act_record] + + return unique_issues + + + def __actOnSite(self, loginbase, site_record): + i_nodes_actedon = 0 + i_nodes_emailed = 0 + + act_recordlist = [] - # Send email, perform node action - # TODO: only send one email per site for a given problem... - if len(act_recordlist) > 0: - act_record = act_recordlist[0] + for nodename in site_record['nodes'].keys(): + diag_record = site_record['nodes'][nodename] + act_record = self.__actOnNode(diag_record) + #print "nodename: %s %s" % (nodename, act_record) + if act_record is not None: + act_recordlist += [act_record] + + unique_issues = self.get_unique_issues(act_recordlist) + + for issue in unique_issues.keys(): + print "\tworking on issue: %s" % issue + issue_record_list = unique_issues[issue] + email_args = self.get_email_args(issue_record_list, loginbase) + + # for each record. + for act_record in issue_record_list: + # if there's a pcu record and email config is set + if 'email_pcu' in act_record: + if act_record['message'] != None and act_record['email_pcu'] and site_record['config']['email']: + # and 'reboot_node' in act_record['stage']: + + email_args['hostname'] = act_record['nodename'] + ticket_id = self.__emailSite(loginbase, + act_record['email'], + emailTxt.mailtxt.pcudown[0], + email_args) + if ticket_id == 0: + # error. + print "got a ticket_id == 0!!!! %s" % act_record['nodename'] + os._exit(1) + pass + email_args['ticket_id'] = ticket_id - # send message before squeezing, b/c - if act_record['message'] != None: - ticket_id = self.__emailSite(loginbase, act_record['email'], - act_record['message'], email_args) + + act_record = issue_record_list[0] + # send message before squeezing + print "\t\tconfig.email: %s and %s" % (act_record['message'] != None, + site_record['config']['email']) + if act_record['message'] != None and site_record['config']['email']: + ticket_id = self.__emailSite(loginbase, act_record['email'], + act_record['message'], email_args) + + if ticket_id == 0: + # error. + print "ticket_id == 0 for %s %s" % (loginbase, act_record['nodename']) + os._exit(1) + pass # Add ticket_id to ALL nodenames - for act_record in act_recordlist: + for act_record in issue_record_list: nodename = act_record['nodename'] # update node record with RT ticket_id - self.act_all[nodename][0]['ticket_id'] = "%s" % ticket_id + if nodename in self.act_all: + self.act_all[nodename][0]['ticket_id'] = "%s" % ticket_id + # if the ticket was previously resolved, reset it to new. + if 'rt' in act_record and \ + 'Status' in act_record['rt'] and \ + act_record['rt']['Status'] == 'resolved': + mailer.setTicketStatus(ticket_id, "new") + status = mailer.getTicketStatus(ticket_id) + self.act_all[nodename][0]['rt'] = status if config.mail: i_nodes_emailed += 1 - # TODO: perform the most severe action? - if b_squeeze: - act_key = act_record['action'] - self.actions[act_key](email_args) + print "\t\tconfig.squeeze: %s and %s" % (config.squeeze, + site_record['config']['squeeze']) + if config.squeeze and site_record['config']['squeeze']: + for act_key in act_record['action']: + self.actions[act_key](email_args) i_nodes_actedon += 1 if config.policysavedb: print "Saving Databases... act_all, diagnose_out" - soltesz.dbDump("act_all", self.act_all) + database.dbDump("act_all", self.act_all) # remove site record from diagnose_out, it's in act_all as done. del self.diagnose_db[loginbase] - soltesz.dbDump("diagnose_out", self.diagnose_db) + database.dbDump("diagnose_out", self.diagnose_db) - print "Hit enter to continue..." - sys.stdout.flush() - line = sys.stdin.readline() + print "sleeping for 1 sec" + time.sleep(1) + #print "Hit enter to continue..." + #sys.stdout.flush() + #line = sys.stdin.readline() return (i_nodes_actedon, i_nodes_emailed) def __actOnNode(self, diag_record): nodename = diag_record['nodename'] message = diag_record['message'] - info = diag_record['info'] act_record = {} act_record.update(diag_record) act_record['nodename'] = nodename act_record['msg_format'] = self._format_diaginfo(diag_record) + print "act_record['stage'] == %s " % act_record['stage'] + + # avoid end records, and nmreset records + # reboot_node_failed, is set below, so don't reboot repeatedly. + + if 'monitor-end-record' not in act_record['stage'] and \ + 'nmreset' not in act_record['stage'] and \ + 'reboot_node_failed' not in act_record: + + if "DOWN" in act_record['log'] and \ + 'pcu_ids' in act_record['plcnode'] and \ + len(act_record['plcnode']['pcu_ids']) > 0: + + print "%s" % act_record['log'], + print "%15s" % (['reboot_node'],) + # Set node to re-install + plc.nodeBootState(act_record['nodename'], "rins") + try: + ret = reboot_node({'hostname': act_record['nodename']}) + except Exception, exc: + print "exception on reboot_node:" + import traceback + print traceback.print_exc() + ret = False + + if ret: # and ( 'reboot_node_failed' not in act_record or act_record['reboot_node_failed'] == False): + # Reboot Succeeded + print "reboot succeeded for %s" % act_record['nodename'] + act_record2 = {} + act_record2.update(act_record) + act_record2['action'] = ['reboot_node'] + act_record2['stage'] = "reboot_node" + act_record2['reboot_node_failed'] = False + act_record2['email_pcu'] = False + + if nodename not in self.act_all: + self.act_all[nodename] = [] + print "inserting 'reboot_node' record into act_all" + self.act_all[nodename].insert(0,act_record2) + + # return None to avoid further action + print "Taking no further action" + return None + else: + print "reboot failed for %s" % act_record['nodename'] + # set email_pcu to also send pcu notice for this record. + act_record['reboot_node_failed'] = True + act_record['email_pcu'] = True - print "%s" % act_record['log'], - print "%15s" % act_record['action'] + print "%s" % act_record['log'], + print "%15s" % act_record['action'] - if act_record['stage'] is not 'monitor-end-record': + if act_record['stage'] is not 'monitor-end-record' and \ + act_record['stage'] is not 'nmreset': if nodename not in self.act_all: self.act_all[nodename] = [] @@ -910,7 +1294,7 @@ class Action(Thread): return act_record def analyseSites(self): - i_sites = 0 + i_sites_observed = 0 i_sites_diagnosed = 0 i_nodes_diagnosed = 0 i_nodes_actedon = 0 @@ -921,19 +1305,20 @@ class Action(Thread): sorted_sites.sort() for loginbase in sorted_sites: site_record = self.sickdb[loginbase] + print "sites: %s" % loginbase i_nodes_diagnosed += len(site_record.keys()) i_sites_diagnosed += 1 (na,ne) = self.__actOnSite(loginbase, site_record) - i_sites += 1 + i_sites_observed += 1 i_nodes_actedon += na i_sites_emailed += ne l_allsites += [loginbase] - return {'sites': i_sites, + return {'sites_observed': i_sites_observed, 'sites_diagnosed': i_sites_diagnosed, 'nodes_diagnosed': i_nodes_diagnosed, 'sites_emailed': i_sites_emailed, @@ -981,38 +1366,6 @@ class Action(Thread): # except Exception, err: # logger.info("POLICY: Problem with DAT, %s" %err) - """ - Returns number of up nodes as the total number *NOT* in act_all with a - stage other than 'steady-state' . - """ - def currentUpAtSite(self, loginbase): - allsitenodes = plc.getSiteNodes(loginbase) - if len(allsitenodes) == 0: - logger.info("Site has no nodes or not in DB") - print "Site has no nodes or not in DB" - return - - numnodes = len(allsitenodes) - sicknodes = [] - # Get all sick nodes at this site - up = 0 - down = 0 - for node in allsitenodes: - - nodename = node - if nodename in self.act_all: # [nodename]: - rec = self.act_all[nodename][0] - if rec['stage'] != "steady-state": - down += 1 - else: - up += 1 - else: - up += 1 - - if up + down != numnodes: - print "ERROR: %s total nodes up and down != %d" % (loginbase, numnodes) - - return up #class Policy(Thread):