frequent crashes / restarts with unbound 1.9.0

[Georg Gottleuber]

Hello unbound users,

I am new to unbound and running an open resolver (a project of
digitalcourage.de against censorship attempts in germany[1]; we have
taken measures outside of unbound against DNS amplification).

It seems that unbound is frequently crashing / restarting. Once the
service is started, nothing indicates a crash, but average time.up (from
stats)never even reaches 120s (our monitoring interval). RAM usage stays
very low and caches are not populated.

After removing all "experimental" features from our config (see below)
the situation got better. Now sometimes time.up reaches 1000s. At night
time.up goes up to higher values, so I think server load plays some role
in this strange behavior. Any ideas?

Platform details:
* Debian 10 (stable), Kernel 4.19.0-10-amd64
* unbound 1.9.0-2+deb10u2
* iptables, limiting UDP output (on public NS-interface)
Hardware & stats:
* VM with 4 VCPUs, 6 GB RAM
* num.query.tls: 60/sec
* total.num.queries: 1500/sec
* load average: 2,67, 3,27, 3,33

-- config --------------------------------------------------------------

server:
	interface: 46.182.19.48 at 53
	interface: 2a02:2970:1002::18 at 53

	interface: 46.182.19.48 at 853
	interface: 2a02:2970:1002::18 at 853
	tls-additional-port: 853
	tls-service-key: /path/dns2.digitalcourage.de/privkey.pem
	tls-service-pem: /path/dns2.digitalcourage.de/fullchain.pem

	outgoing-interface: [some IPv4]

	hide-identity: yes
	hide-version: yes

	prefetch: no

	# public access
	access-control: 0.0.0.0/0 allow
	access-control: ::/0 allow

        verbosity: 1
	logfile: /var/log/unbound.log
	
	# platform / scaling / performance
	num-threads: 4
	msg-cache-slabs: 4
	rrset-cache-slabs: 4
	infra-cache-slabs: 4
	key-cache-slabs: 4
	msg-cache-size: 1024m
	rrset-cache-size: 2048m
	# with libevent
	outgoing-range: 8192
	num-queries-per-thread: 4096
        incoming-num-tcp: 1000
        outgoing-num-tcp: 1000
	
	root-hints: root.hints
	
	deny-any: yes
	minimal-responses: yes

	statistics-interval: 0
	extended-statistics: yes
	
remote-control:
	control-enable: yes
	control-use-cert: no

------------------------------------------------------------------------

regards
Georg


[1]: https://en.wikipedia.org/wiki/Digitalcourage#Campaigns