Okay, the problem is not gone in current 5.7 (5.7.2 from the official repo).
We restarted one affected system 2 days ago. Since then, I have dumped the shm summary a few times.
Shortly after restart: ``` NOTICE: qm_sums: qm_sums(): count= 159 size= 42000 bytes from core: core/dns_cache.c: dns_cache_mk_rd_entry(1117) ```
After running for 10 hours: ``` NOTICE: qm_sums: qm_sums(): count= 9822 size= 2857152 bytes from core: core/dns_cache.c: dns_cache_mk_rd_entry(1117) ```
After running for 2 days: ``` NOTICE: qm_sums: qm_sums(): count= 39931 size= 12716680 bytes from core: core/dns_cache.c: dns_cache_mk_rd_entry(1117) ```
Answering your question about the amount of different domains: We have a pretty static load sharing over all proxies, so without an outage the same amount of customer trunks always get routed through the same proxy. And there are only 10-20 carriers that actually get routed through those proxies. The only special thing are Telekom CompanyFlex trunks, where every customer has their own dns name, which gets set in `$du` when handling the packet. All other trunks get routed simply via `$ru`. In total there are just about 1k trunks registered through each system.