-
Notifications
You must be signed in to change notification settings - Fork 216
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix: Clean up interfaceLockMap entries on endpoint deletion #1249
base: main
Are you sure you want to change the base?
Changes from 10 commits
08ec2d8
0fb0297
280a6cb
50c2a15
5912634
6b11dd5
1a1187a
3e78e6b
a4d7e14
31aa74c
fdab1aa
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -373,36 +373,44 @@ func (p *packetParser) clean(rtnl nltc, qdisc *tc.Object) { | |
} | ||
|
||
func (p *packetParser) endpointWatcherCallbackFn(obj interface{}) { | ||
// Contract is that we will receive an endpoint event pointer. | ||
event := obj.(*endpoint.EndpointEvent) | ||
if event == nil { | ||
return | ||
} | ||
|
||
iface := event.Obj.(netlink.LinkAttrs) | ||
|
||
ifaceKey := ifaceToKey(iface) | ||
lockMapVal, _ := p.interfaceLockMap.LoadOrStore(ifaceKey, &sync.Mutex{}) | ||
mu := lockMapVal.(*sync.Mutex) | ||
mu.Lock() | ||
defer mu.Unlock() | ||
|
||
switch event.Type { | ||
case endpoint.EndpointCreated: | ||
p.l.Debug("Endpoint created", zap.String("name", iface.Name)) | ||
p.createQdiscAndAttach(iface, Veth) | ||
// Get or create mutex atomically | ||
lockMapVal, loaded := p.interfaceLockMap.LoadOrStore(ifaceKey, &sync.Mutex{}) | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Why bring this inside the switch case? We are duplicating code in L399-L404. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Hey Anubhab, Yeah, I get the concern about code duplication, but in this case I think it makes sense to keep the mutex logic separate in each case. The create and delete paths need different mutex handling - creation needs to make a new mutex if it doesn't exist, while deletion only works with existing ones. So, I added an extra check. Trying to handle it outside the switch would probably make the code more generalized and less clear, in addition we are locking the map if the other cases will be added. What do you think? Happy to revert the code to the previous state, if you think if it's unnecessary |
||
mu := lockMapVal.(*sync.Mutex) | ||
mu.Lock() | ||
defer mu.Unlock() | ||
|
||
// Only proceed with creation if this is a new interface | ||
if !loaded { | ||
p.l.Debug("Endpoint created", zap.String("name", iface.Name)) | ||
p.createQdiscAndAttach(iface, Veth) | ||
} | ||
|
||
case endpoint.EndpointDeleted: | ||
p.l.Debug("Endpoint deleted", zap.String("name", iface.Name)) | ||
// Clean. | ||
lockMapVal, ok := p.interfaceLockMap.Load(ifaceKey) | ||
if !ok { | ||
return | ||
} | ||
mu := lockMapVal.(*sync.Mutex) | ||
mu.Lock() | ||
defer mu.Unlock() | ||
|
||
// Clean up operations | ||
if value, ok := p.tcMap.Load(ifaceKey); ok { | ||
v := value.(*tcValue) | ||
p.clean(v.tc, v.qdisc) | ||
// Delete from map. | ||
p.tcMap.Delete(ifaceKey) | ||
} | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. We need to delete the There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Reverted my latest changes, so I brought back the interfaceLockMap and removing the ifacekey from the map |
||
default: | ||
// Unknown. | ||
p.l.Debug("Unknown event", zap.String("type", event.Type.String())) | ||
p.interfaceLockMap.Delete(ifaceKey) | ||
} | ||
} | ||
|
||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why are you deleting this? The
interfaceLockMap
allows us to store a per interface lock and we can create/delete multiple qdisc in parallel. This is necessary (in place of a single lock) because large number of pods can come up at the same time, and we should start capturing packets as quickly as possible.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm still not clear on this. Why is a per-interface mutex necessary for packetparser to handle concurrent create/delete qdisc operations? As far as I understand, operations on different interfaces shouldn't cause a data race.