app-metrics/victoria-metrics-cluster
VictoriaMetrics cluster (vminsert + vmselect + vmstorage)
-
victoria-metrics-cluster-1.156.0~amd64
View
Download
Browse License: Apache-2.0 Overlay: rato-verlay -
victoria-metrics-cluster-1.143.0~amd64
View
Download
Browse License: Apache-2.0 Overlay: rato-verlay
ChangeLog
commit e59a2c15a3928b419e2ff33a42488c3a71592e44
Author: ami-gentoo daily-update (i-023fd91565f88f152) <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 20:42:30 2026 +0000
daily-update: bump overlay packages (daily-20261110-202827)
Source branch: dev
Updated packages:
(none)
Validated (emerged + smoke-tested) packages:
- app-metrics/victoria-metrics-1.156.0
- app-metrics/victoria-metrics-cluster-1.156.0
- dev-util/kiro-bin-1.4.51
- dev-util/kiro-bin-1.4.62
- dev-util/kiro-bin-1.4.86
- dev-util/kiro-bin-1.4.121
- dev-util/kiro-bin-1.4.154-r1
- dev-util/kiro-cli-bin-3.1.0
- dev-util/kirocrew-bin-0.12.1
commit 164a8837ee88b9097cb6035687cbb21c61018019
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 20:59:45 2026 +0100
app-metrics/victoria-metrics{,-cluster}: 1.156.0 needs Go >= 1.27.3
The daily bump of 2026-11-10 copied the 1.143.0 ebuilds, which declare >=dev-lang/go-1.26.3. go.mod of v1.156.0 and v1.156.0-cluster says go 1.27.3, so on a host that already has Go 1.26 Portage would not upgrade Go and the build would fail. Raise the floor; the unchanged copies built on the daily instance only because it pulled the newest Go anyway.
Verified in a fresh Gentoo stage3 container (spyro): emerge of both 1.156.0 source builds with this floor, version output 1.156.0-gentoo; a started victoria-metrics answers /health OK, vminsert/vmselect/vmstorage -version report 1.156.0.
commit 9d65bcae34df37ad839d845a7f1c42eff4e1066e
Author: ami-gentoo daily-update (i-07d602a883afd80de) <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 19:55:21 2026 +0000
daily-update: bump overlay packages (daily-20261110-193927)
Source branch: test
Updated packages:
- dev-util/kiro-bin: 1.4.121 → 1.4.154
- app-metrics/victoria-metrics: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-bin: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-cluster: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-cluster-bin: 1.143.0 → 1.156.0
- net-fs/amazon-efs-utils-bin: 3.1.1 → 3.3.5
Validated (emerged + smoke-tested) packages:
- app-metrics/victoria-metrics-1.156.0
- app-metrics/victoria-metrics-bin-1.156.0
- app-metrics/victoria-metrics-cluster-1.156.0
- app-metrics/victoria-metrics-cluster-bin-1.156.0
- dev-util/kiro-bin-1.4.154
- net-fs/amazon-efs-utils-bin-3.3.5
commit ca9cf65a8b17673833029125c60d8a36748a6328
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Fri Jun 5 20:40:29 2026 +0200
sanitize(victoria-metrics): drop private ansible playbook references
The public overlay must not name the private ansible-playbooks-gentoo
repo. The victoria-metrics pkg_postinst elog (which lands in the
operator email via the portage-elog forwarder) told users to "Tune
retention/listen/memory via the ansible victoriametrics role
(roles/victoriametrics in ansible-playbooks-gentoo)" - leaking the
private repo name and its internal layout into a public ebuild.
Replace with playbook-agnostic guidance: tune via a systemd drop-in
at /etc/systemd/system/victoria-metrics.service.d/ (standard,
public systemd knowledge; no private references).
Cleaned all five touch points across both the single-node and the
(hard-masked) cluster variants:
- victoria-metrics{,-bin} pkg_postinst elog
- victoria-metrics-bin src_install layout comment
- victoria-metrics-cluster{,-bin} src_install layout comment
- both files/victoria-metrics.service ExecStart comments
Overlay-wide grep for ansible/playbook now returns nothing. elog/
comment/service-comment changes do not touch ebuild metadata
variables, so md5-cache is unaffected. bash -n clean on all four
ebuilds.
commit a28359db845ed95ad85425a0d7f85c57461c2249
Author: ami-gentoo weekly-verify (i-02e6a45b6d083d8f5) <16882282+RaminTorabi@users.noreply.github.com>
Date: Sat May 30 18:41:26 2026 +0000
weekly-verify: stamp 1 ebuilds (2026-05-30)
Stamped:
- app-metrics/victoria-metrics-cluster-1.143.0
commit f55501a16bfd8d033ad935bcb6b0329c9573fa1f
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Sun May 24 08:43:40 2026 +0200
feat(app-metrics): add victoria-metrics-cluster{,-bin} (hard-masked)
VictoriaMetrics cluster mode is the horizontally-scalable variant of
the database, decomposing the single-node binary into three services
(vminsert, vmselect, vmstorage) that communicate over RPC and follow
a shared-nothing architecture.
Two new sibling packages mirror the single-node convention:
- app-metrics/victoria-metrics-cluster: source build from the
parallel v$-cluster upstream tag (vendor/ ships, fully
offline build, -mod=vendor).
- app-metrics/victoria-metrics-cluster-bin: prebuilt repackaging of
the upstream victoria-metrics-linux-amd64-v$-cluster.tar.gz
release artifact (three statically-linked Go binaries, install
completes in seconds).
Both ship three systemd unit files (vmstorage.service,
vminsert.service, vmselect.service) with sane single-host defaults:
loopback storageNode wiring, separate http listen ports (8480/8481/
8482), separate RPC ports (8400/8401), data path under
/var/lib/victoria-metrics/cluster/ to avoid collision with the
single-node sibling's /var/lib/victoria-metrics/ root.
Both reuse the existing acct-{user,group}/victoria-metrics service
account. RDEPEND uses an git -P add app-metrics/victoria-metrics-cluster app-metrics/victoria-metrics-cluster-bin metadata/md5-cache/app-metrics/victoria-metrics-cluster-1.143.0 metadata/md5-cache/app-metrics/victoria-metrics-cluster-bin-1.143.0 profiles/package.mask README.md && git -P status -s blocker between cluster and cluster-bin
(same as the single-node pair) but intentionally NOT against the
single-node siblings - the binary names and unit names are
non-overlapping, so Portage doesn't need to enforce uniqueness.
Both packages are hard-masked in profiles/package.mask. Cluster
mode is shipped for completeness and to give a clean upgrade path
if a fleet ever outgrows single-node, but the single-node sibling
is the recommended default for any deployment below ~1M samples/sec
(simpler to operate, ~3x less memory).
Author: ami-gentoo daily-update (i-023fd91565f88f152) <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 20:42:30 2026 +0000
daily-update: bump overlay packages (daily-20261110-202827)
Source branch: dev
Updated packages:
(none)
Validated (emerged + smoke-tested) packages:
- app-metrics/victoria-metrics-1.156.0
- app-metrics/victoria-metrics-cluster-1.156.0
- dev-util/kiro-bin-1.4.51
- dev-util/kiro-bin-1.4.62
- dev-util/kiro-bin-1.4.86
- dev-util/kiro-bin-1.4.121
- dev-util/kiro-bin-1.4.154-r1
- dev-util/kiro-cli-bin-3.1.0
- dev-util/kirocrew-bin-0.12.1
commit 164a8837ee88b9097cb6035687cbb21c61018019
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 20:59:45 2026 +0100
app-metrics/victoria-metrics{,-cluster}: 1.156.0 needs Go >= 1.27.3
The daily bump of 2026-11-10 copied the 1.143.0 ebuilds, which declare >=dev-lang/go-1.26.3. go.mod of v1.156.0 and v1.156.0-cluster says go 1.27.3, so on a host that already has Go 1.26 Portage would not upgrade Go and the build would fail. Raise the floor; the unchanged copies built on the daily instance only because it pulled the newest Go anyway.
Verified in a fresh Gentoo stage3 container (spyro): emerge of both 1.156.0 source builds with this floor, version output 1.156.0-gentoo; a started victoria-metrics answers /health OK, vminsert/vmselect/vmstorage -version report 1.156.0.
commit 9d65bcae34df37ad839d845a7f1c42eff4e1066e
Author: ami-gentoo daily-update (i-07d602a883afd80de) <16882282+RaminTorabi@users.noreply.github.com>
Date: Tue Nov 10 19:55:21 2026 +0000
daily-update: bump overlay packages (daily-20261110-193927)
Source branch: test
Updated packages:
- dev-util/kiro-bin: 1.4.121 → 1.4.154
- app-metrics/victoria-metrics: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-bin: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-cluster: 1.143.0 → 1.156.0
- app-metrics/victoria-metrics-cluster-bin: 1.143.0 → 1.156.0
- net-fs/amazon-efs-utils-bin: 3.1.1 → 3.3.5
Validated (emerged + smoke-tested) packages:
- app-metrics/victoria-metrics-1.156.0
- app-metrics/victoria-metrics-bin-1.156.0
- app-metrics/victoria-metrics-cluster-1.156.0
- app-metrics/victoria-metrics-cluster-bin-1.156.0
- dev-util/kiro-bin-1.4.154
- net-fs/amazon-efs-utils-bin-3.3.5
commit ca9cf65a8b17673833029125c60d8a36748a6328
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Fri Jun 5 20:40:29 2026 +0200
sanitize(victoria-metrics): drop private ansible playbook references
The public overlay must not name the private ansible-playbooks-gentoo
repo. The victoria-metrics pkg_postinst elog (which lands in the
operator email via the portage-elog forwarder) told users to "Tune
retention/listen/memory via the ansible victoriametrics role
(roles/victoriametrics in ansible-playbooks-gentoo)" - leaking the
private repo name and its internal layout into a public ebuild.
Replace with playbook-agnostic guidance: tune via a systemd drop-in
at /etc/systemd/system/victoria-metrics.service.d/ (standard,
public systemd knowledge; no private references).
Cleaned all five touch points across both the single-node and the
(hard-masked) cluster variants:
- victoria-metrics{,-bin} pkg_postinst elog
- victoria-metrics-bin src_install layout comment
- victoria-metrics-cluster{,-bin} src_install layout comment
- both files/victoria-metrics.service ExecStart comments
Overlay-wide grep for ansible/playbook now returns nothing. elog/
comment/service-comment changes do not touch ebuild metadata
variables, so md5-cache is unaffected. bash -n clean on all four
ebuilds.
commit a28359db845ed95ad85425a0d7f85c57461c2249
Author: ami-gentoo weekly-verify (i-02e6a45b6d083d8f5) <16882282+RaminTorabi@users.noreply.github.com>
Date: Sat May 30 18:41:26 2026 +0000
weekly-verify: stamp 1 ebuilds (2026-05-30)
Stamped:
- app-metrics/victoria-metrics-cluster-1.143.0
commit f55501a16bfd8d033ad935bcb6b0329c9573fa1f
Author: Ramin Torabi <16882282+RaminTorabi@users.noreply.github.com>
Date: Sun May 24 08:43:40 2026 +0200
feat(app-metrics): add victoria-metrics-cluster{,-bin} (hard-masked)
VictoriaMetrics cluster mode is the horizontally-scalable variant of
the database, decomposing the single-node binary into three services
(vminsert, vmselect, vmstorage) that communicate over RPC and follow
a shared-nothing architecture.
Two new sibling packages mirror the single-node convention:
- app-metrics/victoria-metrics-cluster: source build from the
parallel v$-cluster upstream tag (vendor/ ships, fully
offline build, -mod=vendor).
- app-metrics/victoria-metrics-cluster-bin: prebuilt repackaging of
the upstream victoria-metrics-linux-amd64-v$-cluster.tar.gz
release artifact (three statically-linked Go binaries, install
completes in seconds).
Both ship three systemd unit files (vmstorage.service,
vminsert.service, vmselect.service) with sane single-host defaults:
loopback storageNode wiring, separate http listen ports (8480/8481/
8482), separate RPC ports (8400/8401), data path under
/var/lib/victoria-metrics/cluster/ to avoid collision with the
single-node sibling's /var/lib/victoria-metrics/ root.
Both reuse the existing acct-{user,group}/victoria-metrics service
account. RDEPEND uses an git -P add app-metrics/victoria-metrics-cluster app-metrics/victoria-metrics-cluster-bin metadata/md5-cache/app-metrics/victoria-metrics-cluster-1.143.0 metadata/md5-cache/app-metrics/victoria-metrics-cluster-bin-1.143.0 profiles/package.mask README.md && git -P status -s blocker between cluster and cluster-bin
(same as the single-node pair) but intentionally NOT against the
single-node siblings - the binary names and unit names are
non-overlapping, so Portage doesn't need to enforce uniqueness.
Both packages are hard-masked in profiles/package.mask. Cluster
mode is shipped for completeness and to give a clean upgrade path
if a fleet ever outgrows single-node, but the single-node sibling
is the recommended default for any deployment below ~1M samples/sec
(simpler to operate, ~3x less memory).

