Bug #4474
closed5.0.x: dns: high resource usage on long lived dns connections
Description
The transaction handling is off for unidirectional protocols like DNS on long lived connections. That is a single TCP/UDP session that receives multiple DNS requests/responses. For DNS this is rare, but some devices do this.
This results in the transaction cleanup trying to remove transactions that have already been removed.
Updated by Victor Julien almost 3 years ago
- Copied from Bug #4437: dns: high resource usage on long lived dns connections added
Updated by Victor Julien almost 3 years ago
Since the fixes in master and 6.x aren't going to be easily backported, we might need a special case workaround for dns in 5.
Updated by Jason Ish almost 3 years ago
In 6.0.x we saw a case where the last cleaned up (free'd transaction) wasn't being recorded correctly. So for every transaction that needed to be cleaned up in 'AppLayerParserTransactionsCleanup', it would try to clean up the 32 preceding transactions that were already cleaned up. So in 'AppLayerParserTransactionsCleanup' we were looping at 32 times to cleanup a single transaction.
With 5.0.x that tracking seems to be OK. We only loop once in cleanup.
I'm wondering if its just general transaction handling improvements between 5 and 6 that are making 6 (with this fix applied) better than 5.0?
Updated by Jason Ish almost 3 years ago
- Status changed from New to In Review
Updated by Jason Ish almost 3 years ago
- Subject changed from dns: high resource usage on long lived dns connections to 5.0.x: dns: high resource usage on long lived dns connections
- Status changed from In Review to Rejected
So far I don't think we have sufficient evidence that this is a real issue in 5.0. 6.0.x is more efficient with respect to transaction handling than 5.0, but this specific issue was an issue introduced in 6.0 that does not show in 5.0, and it was the upgrade to 6.0 where users started to notice an issue.
Rejecting for now.