Repository navigation
Heron v0.8.1
[0.8.1] — 2026-09-12
v0.8.0 was tagged from this same content but never released: the
aarch64-unknown-linux-musl leg of the release matrix failed to link, so no
release was created and nothing shipped. 0.8.1 is that content plus the build
fix — there is no released 0.8.0 to upgrade from.
Changed
-
The aglake storage backend now targets aglake 0.3 and is named for it.
The log platform Heron writes to renamed itself from sglog to Netis Aglake,
and made the rename a breaking boundary rather than an alias layer — the REST
namespace Heron managed index retention through answers410 Gone. Both
management calls move to the native/api/v1/admin/indexes, which is a
capability gain as much as a port: the Splunk-compatible face they used
before is mounted only when the daemon is started with vendored frontend
assets, so an ingest-only deployment could not be told about retention at all
(a released 0.3 tarball reportssplunk_face: false). The crate, config
table and backend value are renamed to match. Older daemons are no longer
supported — retention pushes against one fail with a 404 that says so.
Upstream has since renumbered that release line to 1.5 without breaking the
API again; the backend's live suite passes against both a 0.3 and a 1.5
build.Existing config files keep working:
backend = "sglake"is normalized at
load,[storage.sglake]is accepted as an alias (covering the matching
TS__STORAGE__SGLAKE__*overrides — note the doubled separator after the
prefix), andheron aglake-propsstill answers tosglake-props. Loading an
old spelling raises a warning, not an error; having both table names in one
file is aduplicate fielderror rather than a silent precedence rule.Two environment variables outside the repo keep their old names for the same
reason:SGLAKE_DATA_DIR(read by the staging longevity soak from a
hand-maintained environment file) andSGLOGD_BIN(the cross-backend
differential harness). Renaming those with no fallback would have silently
dropped an invariant rather than failed.
Added
-
Session authentication for aglake's
/api/v1/*faces. aglake 0.3 put a
local user catalog in front of search as well as administration; with one
configured, an unauthenticated Heron gets401on every read.
storage.aglake.username/passwordare exchanged for a session at first
use, andsession_tokenpresents one directly.hec_tokendoes not cover
this — it authenticates ingest only, which is the mistake the new error
messages are written to catch. Sessions are held in the daemon's memory with
a 12-hour TTL and are lost when it restarts, so a401is an expected event:
Heron logs in again and retries once, which absorbs an expired session while
still reporting credentials that are simply wrong.Configuring credentials also lifts the non-loopback warning on
storage.aglake.url— that warning exists because the port was the only
thing standing in front of stored request and response bodies, which is no
longer true once the daemon checks who is asking.
Fixed
- The aarch64 Linux release binary could not be linked. rustc stable passes
-Wl,--fix-cortex-a53-843419for aarch64 targets and zig's linker rejects
unknown arguments outright, so that leg of the release matrix failed and, with
it, the whole release.cargo-zigbuildgained a filter for the flag in
0.23.0; the pin here was 0.19.6, from 2024. Nothing in Heron changed —
rust-toolchain@stablemoved underneath a pinned build tool, which is what
brokev0.8.0. /api/runtime-configwas handing out stored credentials. It returns the
wholeAppConfigto the console, andheron config validate --jsonprints
it, sostorage.aglake.hec_tokenandstorage.clickhouse.passwordwere
readable by anyone who could reach either — on a default deploy the API
listens on0.0.0.0with no authentication of its own. All credential
fields, including the two added in this release, are now excluded from
serialization; loading is unaffected. The console never read them. Anyone
running an earlier version with a HEC token or ClickHouse password
configured should treat that credential as disclosed to its network and
rotate it.- Every documented
TS_*environment override was silently ignored. The
config loader sets__as the path separator without setting a prefix
separator, and the library defaults the latter to the former — so the real
prefix isTS__, and the documentedTS_API__PORT=9090form matched no
prefix and was dropped. Only the documentation is corrected here (to
TS__API__PORT), pinned by a test: changing the separator would alter how
every key is read, which does not belong in a release. - The live suite's self-started aglaked could not start against a current
build. From0.3.0.2653on, aglaked binds six dedicated receivers on fixed
0.0.0.0ports by default (HEC 8088, OTLP 4318/4317, syslog 514, S2S 9997,
ES-compat 9200) and exits if any cannot bind — 514 is privileged, and the
rest collide with anything else on the host. The two tests that own their
daemon now switch those off and move the dedicated HEC listener to loopback,
probing--helpfirst so they still work against a build predating the
flags. Note--hec-http-enabled falseis not the way to do it: it also
disables the HEC input on--listen, which is the one Heron writes
through.
What's Changed
- ci(release): pin both halves of the zig toolchain, and cut v0.8.1 by @vaderyang in #199
Full Changelog: v0.8.0...v0.8.1