135/netdisco
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fixes for race conditions and dupes in job queue

Browse Source 
we had situations where the manager would start workers on the same job,
either because of race conditions or because at the time of queueing it wasn't
known that the jobs were targeting the same device (due to device aliases).

this commit removes duplicate jobs, reduces the need for locking on the job
queue, and makes use of lldpRemChassisId to try to deduplicate jobs before
they are started. in effect we have several goes to prevent duplicate jobs:

1. at neighbor discovery time we try to skip queueing same lldpRemChassisId
2. at job selection we 'error out' jobs with same profile as job selected
3. at job selection we check for running job with same profile as selected
4. the job manager process also checks for duplicate job profiles
5. at job lock we abort if the job was 'errored out'

all together this seems to work well. a test on a large university network of
303 devices (four core routers and the rest edge routers, runing VRF with many
duplicate identities), ~1200 subnets, ~50k hosts, resulted in no DB deadlock
or contention and a complete discover+arpnip+macsuck (909 jobs) in ~3 minutes
(with ~150 duplicate jobs identified and skipped).
This commit is contained in:
Oliver Gorwits
2017-11-23 19:23:55 +00:00
parent 1bbe8c9164
commit 0bb15f36b9
10 changed files with 166 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
lib/App/Netdisco/Worker/Plugin/Discover/CanonicalIP.pm
View File

@@ -72,6 +72,10 @@ register_worker({ phase => 'main', driver => 'snmp' }, sub {
$device->renumber($new_ip)
or die "cannot renumber to: $new_ip"; # rollback
# is not done in renumber but required otherwise confusing at job end!
schema('netdisco')->resultset('Admin')
->find({job => $job->id})->update({device => $new_ip});
return Status->noop(sprintf ' [%s] device - changed IP to %s (%s)',
$old_ip, $device->ip, ($device->dns || ''));
});

12
lib/App/Netdisco/Worker/Plugin/Discover/Neighbors.pm
View File

@@ -39,11 +39,13 @@ register_worker({ phase => 'main', driver => 'snmp' }, sub {
or return Status->defer("discover failed: could not SNMP connect to $device");
my @to_discover = store_neighbors($device);
my %seen_id = ();
# only enqueue if device is not already discovered,
# discover_* config permits the discovery
foreach my $neighbor (@to_discover) {
my ($ip, $remote_type) = @$neighbor;
my ($ip, $remote_type, $remote_id) = @$neighbor;
next if $remote_id and $seen_id{ $remote_id }++;
my $device = get_device($ip);
next if $device->in_storage;
@@ -55,10 +57,14 @@ register_worker({ phase => 'main', driver => 'snmp' }, sub {
next;
}
# risk of things going wrong...?
# https://quickview.cloudapps.cisco.com/quickview/bug/CSCur12254
jq_insert({
device => $ip,
action => 'discover',
subaction => 'with-nodes',
($remote_id ? (device_key => $remote_id) : ()),
});
}
});
@@ -171,7 +177,7 @@ sub store_neighbors {
# useable remote IP...
if ($remote_ip eq '0.0.0.0' or
check_acl_no($remote_ip, 'group:__LOCAL_ADDRESSES__')) {
check_acl_no($remote_ip, 'group:__LOCAL_ADDRESSES__')) {
if ($remote_id) {
my $devices = schema('netdisco')->resultset('Device');
@@ -228,7 +234,7 @@ sub store_neighbors {
debug sprintf
' [%s] neigh - adding neighbor %s, type [%s], on %s to discovery queue',
$device->ip, $remote_ip, ($remote_type || ''), $port;
push @to_discover, [$remote_ip, $remote_type];
push @to_discover, [$remote_ip, $remote_type, $remote_id];
$remote_port = $c_port->{$entry};
if (defined $remote_port) {