24 Jul
2017
24 Jul
'17
9:12 a.m.
Module: check_mk
Branch: master
Commit: f81a29332490829b82d308a135ccdabd7a6f3df2
URL:
http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=f81a2933249082…
Author: Andreas Boesl <ab(a)mathias-kettner.de>
Date: Mon Jul 24 10:51:39 2017 +0200
4852 FIX Livestatus connections: fixed bug where data from previous connection got reused
When using persistent connections, there was a risk that the response data from
previous livestatus queries got intermingled with the data from the latest query.
This caused quite a lot of confusing for most of the GUI plugins, which crashed
occasionally.
Change-Id: Ib0629b08c1313a058f73339c60eb807ee9dc0ca6
---
.werks/4852 | 16 ++++++++++++++++
livestatus/api/python/livestatus.py | 30 +++++++++++++++++++-----------
2 files changed, 35 insertions(+), 11 deletions(-)
diff --git a/.werks/4852 b/.werks/4852
new file mode 100644
index 0000000..294ec72
--- /dev/null
+++ b/.werks/4852
@@ -0,0 +1,16 @@
+Title: Livestatus connections: fixed bug where data from previous connection got reused
+Level: 1
+Component: livestatus
+Class: fix
+Compatible: compat
+Edition: cre
+State: unknown
+Version: 1.5.0i1
+Date: 1500885870
+
+When using persistent connections, there was a risk that the response data from
+previous livestatus queries got intermingled with the data from the latest query.
+
+
+This caused quite a lot of confusing for most of the GUI plugins, which crashed
occasionally.
+
diff --git a/livestatus/api/python/livestatus.py b/livestatus/api/python/livestatus.py
index 5df2c38..7098da2 100644
--- a/livestatus/api/python/livestatus.py
+++ b/livestatus/api/python/livestatus.py
@@ -315,7 +315,10 @@ class BaseConnection:
def disconnect(self):
self.socket = None
if self.persist:
- del persistent_connections[self.socketurl]
+ try:
+ del persistent_connections[self.socketurl]
+ except KeyError:
+ pass
def receive_data(self, size):
result = ""
@@ -384,6 +387,7 @@ class BaseConnection:
try:
length = int(resp[4:15].lstrip())
except:
+ self.disconnect()
raise MKLivestatusSocketError("Malformed output. Livestatus TCP
socket might be unreachable.")
data = self.receive_data(length)
@@ -392,6 +396,7 @@ class BaseConnection:
try:
return ast.literal_eval(data)
except:
+ self.disconnect()
raise MKLivestatusSocketError("Malformed output")
elif code == "404":
@@ -424,6 +429,10 @@ class BaseConnection:
raise
except Exception, e:
+ # Catches
+ # MKLivestatusQueryError
+ # MKLivestatusSocketError
+ # FIXME: ? self.disconnect()
raise MKLivestatusSocketError("Unhandled exception: %s" % e)
@@ -722,6 +731,7 @@ class MultiSiteConnection(Helpers):
result += r
stillalive.append( (sitename, site, connection) )
except Exception, e:
+ connection.disconnect()
self.deadsites[sitename] = {
"exception" : e,
"site" : site,
@@ -733,13 +743,15 @@ class MultiSiteConnection(Helpers):
# of Limit: since all sites are queried in parallel, the Limit: is simply
# applied to all sites - resulting in possibly more results then Limit requests.
def query_parallel(self, query, add_headers = ""):
+ stillalive = []
if self.only_sites != None:
- active_sites = [ c for c in self.connections if c[0] in self.only_sites ]
+ connect_to_sites = [ c for c in self.connections if c[0] in self.only_sites
]
+ # Unused sites are assumed to be alive
+ stillalive.extend( [ c for c in self.connections if c[0] not in
self.only_sites])
else:
- active_sites = self.connections
+ connect_to_sites = self.connections
start_time = time.time()
- stillalive = []
limit = self.limit
if limit != None:
limit_header = "Limit: %d\n" % limit
@@ -747,7 +759,7 @@ class MultiSiteConnection(Helpers):
limit_header = ""
# First send all queries
- for sitename, site, connection in active_sites:
+ for sitename, site, connection in connect_to_sites:
try:
connection.send_query(query, add_headers + limit_header)
except Exception, e:
@@ -764,11 +776,7 @@ class MultiSiteConnection(Helpers):
# Then retrieve all answers. We will be as slow as the slowest of all
# connections.
result = []
- for sitename, site, connection in self.connections:
- if self.only_sites != None and sitename not in self.only_sites:
- stillalive.append( (sitename, site, connection) ) # state unknown, assume
still alive
- continue
-
+ for sitename, site, connection in connect_to_sites:
try:
r = connection.recv_response(query, add_headers + limit_header)
stillalive.append( (sitename, site, connection) )
@@ -780,12 +788,12 @@ class MultiSiteConnection(Helpers):
continue
except Exception, e:
+ connection.disconnect()
self.deadsites[sitename] = {
"exception" : e,
"site" : site,
}
-
self.connections = stillalive
return result