A footprint and DNSSEC release: TelltaleDNS uses much less memory, and DNSSEC validation gets the corrections a live trial called for.
Highlights
- Less memory. A Raspberry Pi 4 with 2.7 million blocked names went from about 83 MiB to about 50 MiB.
- The domain lists are memory-mapped and their pages released once the lookup index answers queries.
- Each cached answer is one allocation instead of three.
- The response buffer for a query sent upstream is allocated when the answer arrives, sized to it. Before, it was a 64 KiB buffer held for the whole wait.
- DNSSEC you can trust in permissive mode. Names in unsigned zones reached through CNAMEs (Netflix, Amazon, BBC) were judged bogus by the validation library. TelltaleDNS now proves for itself that such a zone is unsigned (the parent's signed denial of its DS) and answers it as insecure. Proven zones are remembered for 15 minutes, which also ends a slow lookup loop on some of those names.
- In permissive mode every bogus answer is logged with its name and reason, so a trial shows what
validatewould refuse. - A validation that hits the library's depth limit is treated as indeterminate (answered without AD), not bogus.
- In permissive mode every bogus answer is logged with its name and reason, so a trial shows what
- Fewer failed lookups. When an upstream closes a DoT or TCP connection after a query was sent but before answering, the query is retried once on a fresh connection.
- Quieter logs. The validation library's own warnings are off (
TELLTALE_LOG_DNSSEC_LIBRARY=1turns them back on), and cluster replicas no longer repeat configuration warnings on every replicated change.
Upgrading from 0.1.0
[cache] max_bytesnow counts real memory. It used to undercount each entry by about a third, so the same setting now holds fewer entries. The cache uses about the memory you asked for. Raise the limit if you sized it by entry count.- The query log starts a new segment part after 16,384 distinct names in an hour (was 65,536), which keeps its memory near 1 MiB. Busy hours write more, smaller parts; searches are unchanged.
- Nothing else to change in your configuration.
Known issues
- DNSSEC
validatemode: right after a start, an upstream that closes fresh DoT connections can make the first validations fail, so a signed name may get SERVFAIL until its keys are cached. Keeppermissivefor now and read the log; see ADR-098 and the roadmap (T10.9). _dns.resolver.arpa(encrypted-DNS discovery) is forwarded upstream instead of being answered locally.
For developers
bench/gate.pymeasures the v1.0 gate on one machine;docs/v1-gate.mdtracks every gate.bench/memprobe.pyshows memory by mapping.bench/heapprof.pywith thedhat-heapcargo feature (development only) names the code holding the heap.- The release workflow's version bump now also updates the OpenAPI document.
Full Changelog: v0.1.0...v0.2.0