DoH unreliable on EdgeRouter X (mipsle) router

[BrianG61UK]

When using DoH I find that after some time (typically less than 24 hours) I find I'm unable to resolve any DNS queries and the dnscrypt-proxy process is consuming 100++% CPU. No problem when using DNSCrypt rather than DoH.

A similar problem with nextdns/nextdns cli was fixed by reverting from go 1.17 to the latest go 1.16.

[hugepants]

I have experienced this with the ER-X on OpenWrt. Related issues: #797 and #794

I was not able to solve the problem for DoH after trying different tls_cipher_suite values, so my solution was to use only DNSCrypt resolvers.

That is interesting about reverting Go to 1.16, I don't know if this is feasible in dnscrypt-proxy, but I hope they can identify the root cause.

[BrianG61UK]

Thanks. In particular, thanks for the useful links.

[BrianG61UK]

Nothing seems to be happening about this :(

[lifenjoiner]

I just built 2 versions by go v1.18.3 with github CI, that also have mipsle builds. If you are interested, try from:
TestAndroid-3

Update:
Now, Release 2.1.2 is built with go v1.18.4.

[hugepants]

@BrianG61UK
For me this issue is solved in dnscrypt-proxy 2.1.2 (Go 1.18.4), can you please try it and let us know your results?

And for the best DoH performance on MIPS, you should use a server that supports TLSv1.3 (304) and offers a TLS_CHACHA20_POLY1305_SHA256 (4867) cipher, which many do these days. Cloudflare, Google, NextDNS all do. You can check this by setting log_level = 1 in dnscrypt-proxy.toml.

[BrianG61UK]

@hugepants Okay. I'm now running 2.1.2 and DoH ('cisco-doh') instead of my usual DNSCrypt ('cisco').
I'll see how I get on over the next few days.

[BrianG61UK]

Seems like it must be fixed. No problems and it typically lasted less than a day when the bug was present.

[hugepants]

Excellent, thanks for the update.