vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

Compare commits

base: vinzenz:065933b83ebc1583179055215b2c5e4456745140
Branches Tags
vinzenz:main
..
compare: vinzenz:95b89970eeadd05f129406e3ac2928c699b7cdcf
Branches Tags
vinzenz:main
Author SHA1 Message Date
atlas
95b89970ee swarm-victoriametrics: require 1.153.0, carry the history off the DynamicUser path 
The socket scheme needs VictoriaMetrics 1.153.0 or later: older builds
create the socket 0600 whatever the umask, so nginx could not connect
and every consumer would get 502 with the module evaluating clean.
deploy.victoriametrics.package is an operator option, so an assertion
now refuses an older one, naming the option and the reason.

Turning DynamicUser off moves the state directory from
/var/lib/private/victoriametrics (behind a /var/lib/victoriametrics
symlink) to a plain /var/lib/victoriametrics. A oneshot ordered before
the store moves the old directory there and chowns it to the static
user when the plain path is the old symlink, empty, or missing, and does
nothing once the private directory is gone. Without it the persistent
container's history depends on systemd's own DynamicUser-off migration.

Refs #4815
 2026-10-14 18:15:20 +02:00
atlas
5e653ab9b5 swarm-victoriametrics: own network namespace, gateway over a unix socket 
The metrics store's container gets privateNetwork = true with no
interface but loopback. VictoriaMetrics listens on
unix:/run/swarm-victoriametrics/victoriametrics.sock, a host directory
(root:nginx 0770) bind-mounted into the container, and every vhost
location that forwarded to 127.0.0.1:8428 now forwards to that socket.

The store runs as a static user whose primary group is nginx's gid,
with UMask 0007, so the socket it creates is 0770 group nginx. This
relies on VictoriaMetrics >= 1.153.0 honouring the umask (the locked
nixpkgs ships 1.153.0). nixpkgs' readiness loop curls a TCP URL, so
postStart is replaced with the same loop over --unix-socket.

swarm.victoriametrics.port is removed (nothing binds it any more), with
a mkRemovedOptionModule entry, and dropped from swarm-otel's host
port-collision assertion. The container's resolver unit is dropped:
there is no interface to reach the bridge resolver through, and the
store resolves no names.

The collector scrape, the collector push and Grafana's datasource
already reach the store by its published domain, so they are unchanged.

Refs #4815
Refs #3560
 2026-10-14 18:15:20 +02:00
8 changed files with 244 additions and 59 deletions
Download patch file Download diff file Expand all files Collapse all files

11
docs/swarm/services.md
View file

@ -168,8 +168,10 @@ Grafana start — see [`secrets.md`](secrets.md).
**Where the data comes from.** The swarm's OTEL collector, below.
Neither container is reachable except through the gateway: both bind
loopback, and VictoriaMetrics authenticates nothing itself. The
Neither container is reachable except through the gateway: each listens
on a unix socket only the gateway's nginx can open, and VictoriaMetrics
runs in a network namespace of its own with no interface but loopback.
VictoriaMetrics authenticates nothing itself. The
`metrics.<swarm-domain>` vhost forwards only the store's read API (the
Prometheus query, series, label and metadata routes, plus the vmui web UI)
without a credential. It forwards the collector's OTLP push, and the
@ -206,9 +208,8 @@ swarm.
| `swarm.victorialogs.domain` | Only to rename it. |
| `swarm.victorialogs.port` | Only if something else on the services host already claims `9428`. |
Like the metrics store, it binds loopback and takes no credential of its
own: the gateway vhost is the only way in, and the collector is the only
intended writer.
It binds loopback and takes no credential of its own: the gateway vhost is
the only way in, and the collector is the only intended writer.
The collector also scrapes this store's own metrics, at
`https://<swarm.victorialogs.domain>/metrics`, so a collector on a host

5
nix/host-modules/deploy.nix
View file

@ -80,6 +80,11 @@ in
[ "services" "hyperhive" "swarm" "victoriametrics" "enable" ]
[ "services" "hyperhive" "deploy" "victoriametrics" "enable" ]
)
(lib.mkRemovedOptionModule [ "services" "hyperhive" "swarm" "victoriametrics" "port" ] ''
VictoriaMetrics listens on a unix socket in a network namespace of its
own, and the gateway reaches it there, so it binds no host port. Remove
this definition.
'')
(lib.mkRenamedOptionModule
[ "services" "hyperhive" "swarm" "victorialogs" "enable" ]
[ "services" "hyperhive" "deploy" "victorialogs" "enable" ]

3
nix/host-modules/swarm-otel.nix
View file

@ -757,8 +757,7 @@ in
cfg.producerPort
cfg.storeProducerPort
otelCfg.collector.port
]
++ lib.optional deployCfg.victoriametrics.enable vmCfg.port;
];
all = derived ++ others;
in
lib.length (lib.unique all) == lib.length all;

3
nix/host-modules/swarm-victorialogs.nix
View file

@ -455,8 +455,7 @@ in
enable = true;
package = deployCfg.victorialogs.package;
# ⚠️ PINNED TO LOOPBACK for the same reason the metrics store is,
# and it matters more here: upstream's default listens on every
# ⚠️ PINNED TO LOOPBACK: upstream's default listens on every
# interface, and this endpoint accepts writes as well as reads.
# Nothing authenticates it either — upstream offers one basic-auth
# pair (`-httpAuth.username` / `-httpAuth.password`) and this

21
nix/host-modules/swarm-victoriametrics-service.nix
View file

@ -1,5 +1,5 @@
# The swarm's metrics store as every hive sees it: the name it answers on and
# its port, identical on every host. What the host running it decides, and the
# The swarm's metrics store as every hive sees it: the name it answers on,
# identical on every host. What the host running it decides, and the
# container itself, are in ./swarm-victoriametrics.nix.
{
lib,
@ -16,8 +16,8 @@ let
domainBase = if swarmDomain == null then "invalid" else swarmDomain;
in
{
# What the store IS from any hive's point of view: the name it answers on and
# the port. `enable`, `package` and `retentionPeriod` are decisions of the
# What the store IS from any hive's point of view: the name it answers on.
# `enable`, `package` and `retentionPeriod` are decisions of the
# host that runs it and live under `deploy.*`.
options.services.hyperhive.swarm.victoriametrics = {
machine = lib.mkOption {
@ -42,18 +42,5 @@ in
child decides which CA may sign it.
'';
};
port = lib.mkOption {
type = lib.types.port;
default = 8428;
description = ''
Port VictoriaMetrics listens on, bound to loopback only on the
host that runs it: its write endpoint is unauthenticated, so the
gateway's vhost there is the only route to it. Upstream's own
default, kept so an operator reading VictoriaMetrics
documentation finds what they expect.
'';
};
};
}

172
nix/host-modules/swarm-victoriametrics.nix
View file

@ -18,13 +18,23 @@
let
cfg = config.services.hyperhive.swarm.victoriametrics;
deployCfg = config.services.hyperhive.deploy;
networkCfg = config.services.hyperhive.network;
hyperhiveCfg = config.services.hyperhive;
gatewayCfg = hyperhiveCfg.gateway;
# Shared host netns, like every sibling swarm container: the gateway
# reaches this at 127.0.0.1:<port>.
privateNetwork = false;
# A netns of its own with no interface but loopback: the store makes no
# outbound calls, and the gateway reaches it over `socketPath` below, so
# nothing on the host or the hive bridge can dial its listener.
privateNetwork = true;
# Fixed, not an option: the directory's mode is what keeps the socket
# private, so it holds this one socket and nothing else.
socketDir = "/run/swarm-victoriametrics";
socketPath = "${socketDir}/victoriametrics.sock";
upstream = path: "http://unix:${socketPath}:${path}";
# The gid the gateway's nginx runs as. Only the number crosses into the
# container, which knows it as `gateway-nginx`.
nginxGid = config.ids.gids.nginx;
# The store's read API: the only part of it the vhost serves without a
# credential. Each location's `proxy_pass` carries the path, so the store
@ -64,7 +74,7 @@ let
];
readLocations =
let
to = path: { proxyPass = "http://127.0.0.1:${toString cfg.port}${path}"; };
to = path: { proxyPass = upstream path; };
in
lib.listToAttrs (
map (p: lib.nameValuePair "= ${p}" (to p)) readPaths
@ -72,8 +82,8 @@ let
);
in
{
# What the store IS from any hive's point of view, the name it answers on and
# the port, is `swarm.victoriametrics` in ./swarm-victoriametrics-service.nix.
# What the store IS from any hive's point of view, the name it answers on,
# is `swarm.victoriametrics` in ./swarm-victoriametrics-service.nix.
# `enable`, `package` and `retentionPeriod` are decisions of the host that
# runs it and live under `deploy.*`.
@ -104,6 +114,22 @@ in
};
config = lib.mkIf deployCfg.victoriametrics.enable {
assertions = [
{
assertion = lib.versionAtLeast (lib.getVersion deployCfg.victoriametrics.package) "1.153.0";
message = ''
services.hyperhive.deploy.victoriametrics.package must be
VictoriaMetrics 1.153.0 or later (this one is
${lib.getVersion deployCfg.victoriametrics.package}).
The gateway reaches the store over a unix socket, and only from
1.153.0 does VictoriaMetrics create that socket under the process
umask. An older build creates it 0600 whatever the umask, so nginx
cannot connect and every read, push and scrape fails with 502.
'';
}
];
# The gateway name, inside `deployCfg.victoriametrics.enable` — that
# guard is the load-bearing part. Every hive in a swarm may know this
# store exists, but only the host that RUNS it may claim the name; a
@ -111,8 +137,7 @@ in
# have.
services.hyperhive.gateway.localNames = [ cfg.domain ];
# This host serves the vhost, and the container behind it resolves
# through the hive's dnsmasq.
# This host serves the vhost, and the hive's dnsmasq answers its name.
services.hyperhive.gateway.enable = lib.mkDefault true;
services.hyperhive.gateway.dns.enable = lib.mkDefault true;
@ -138,7 +163,7 @@ in
# follow it and POST its batch at a login page that answers 200 —
# ingest reporting healthy while storing nothing.
"= /opentelemetry/api/v1/push" = {
proxyPass = "http://127.0.0.1:${toString cfg.port}/opentelemetry/api/v1/push";
proxyPass = upstream "/opentelemetry/api/v1/push";
extraConfig = ''
auth_request /__metrics_push_authz;
'';
@ -153,7 +178,7 @@ in
# `publishedScrapeTargets.victoriametrics` in ./swarm-otel-service.nix;
# the two must agree, since that string is also the token's audience.
"= /metrics" = {
proxyPass = "http://127.0.0.1:${toString cfg.port}/metrics";
proxyPass = upstream "/metrics";
extraConfig = ''
auth_request /__metrics_push_authz;
'';
@ -188,6 +213,17 @@ in
};
};
# The socket directory, created host-side before the container starts:
# nixos-container refuses to start when a bind source is missing.
#
# Root-owned and group nginx, `0770`: the store creates its socket here
# through the group (its uid is allocated inside the container, so the
# host cannot name it), and the gateway traverses it the same way.
# Everything else is kept out.
systemd.tmpfiles.rules = [
"d ${socketDir} 0770 root ${toString nginxGid} - -"
];
containers.${cfg.machine} = {
autoStart = true;
ephemeral = false;
@ -196,31 +232,105 @@ in
extraFlags = [ "--link-journal=host" ];
inherit privateNetwork;
config =
{ ... }:
{
imports = [
../container-modules/swarm-container.nix
(import ../container-modules/swarm-container-resolver.nix {
inherit (networkCfg) bridgeIp;
dnsConsumers = [ "victoriametrics.service" ];
})
];
# A bind mount rather than a path both sides happen to know: a
# container's `/run` is an nspawn tmpfs the host cannot see into.
bindMounts.${socketDir} = {
hostPath = socketDir;
isReadOnly = false;
};
services.hyperhive.swarmContainer = { inherit privateNetwork; };
config =
{ config, pkgs, ... }:
{
imports = [ ../container-modules/swarm-container.nix ];
# No resolver unit: with no interface but loopback there is no
# nameserver to reach, and the store resolves no names.
services.hyperhive.swarmContainer = {
inherit privateNetwork;
writesOwnResolvConf = false;
};
# The socket's group is the store's primary group, so that group is
# nginx's gid. A static user, not nixpkgs' DynamicUser: a dynamic
# user's group is allocated per start, and nginx is never in it.
users.groups.gateway-nginx.gid = nginxGid;
users.users.victoriametrics = {
isSystemUser = true;
group = "gateway-nginx";
};
systemd.services.victoriametrics = {
serviceConfig = {
DynamicUser = lib.mkForce false;
User = "victoriametrics";
Group = "gateway-nginx";
# The store creates its socket under the process umask, so
# `0007` makes it `0770` and nginx's group can connect.
UMask = "0007";
};
# nixpkgs' readiness loop curls `http://<listenAddress>/ping`,
# which no `unix:` address answers.
postStart = lib.mkForce ''
until ${lib.getExe pkgs.curl} -s -o /dev/null --unix-socket ${socketPath} http://localhost/ping; do
sleep 1;
done
'';
};
# Moves the store's history out of the DynamicUser layout, where
# `/var/lib/<stateDir>` is a symlink into `/var/lib/private/`, to the
# plain directory the static user's StateDirectory is. Runs before
# the store on every start and does nothing once
# `/var/lib/private/<stateDir>` is gone. When both paths hold data
# it moves neither and the store starts on the plain one.
systemd.services.victoriametrics-state-migrate =
let
old = "/var/lib/private/${config.services.victoriametrics.stateDir}";
new = "/var/lib/${config.services.victoriametrics.stateDir}";
in
{
description = "move VictoriaMetrics' DynamicUser state directory to its static path";
before = [ "victoriametrics.service" ];
requiredBy = [ "victoriametrics.service" ];
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
SyslogIdentifier = "victoriametrics-state-migrate";
};
path = [ pkgs.coreutils ];
script = ''
set -eu
old=${lib.escapeShellArg old}
new=${lib.escapeShellArg new}
if [ ! -d "$old" ] || [ -L "$old" ]; then
exit 0
fi
if [ -L "$new" ]; then
rm "$new"
elif [ -d "$new" ]; then
if [ -n "$(ls -A "$new")" ]; then
echo "victoriametrics-state-migrate: $old and $new both hold data; moving neither" >&2
exit 0
fi
rmdir "$new"
fi
mv "$old" "$new"
chown -R victoriametrics:gateway-nginx "$new"
echo "victoriametrics-state-migrate: moved $old to $new"
'';
};
services.victoriametrics = {
enable = true;
package = deployCfg.victoriametrics.package;
retentionPeriod = deployCfg.victoriametrics.retentionPeriod;
# ⚠️ PINNED TO LOOPBACK, and this is a correction rather than a
# preference: upstream's default is `:8428`, i.e. every
# interface. The gateway is the only intended client and it is on
# this host, so binding wider would publish an unauthenticated
# write endpoint (see the OTLP note below) to whatever the host
# is reachable on.
listenAddress = "127.0.0.1:${toString cfg.port}";
# ⚠️ A socket, not upstream's default `:8428` on every interface:
# the gateway is the only intended client, and this listener
# carries an unauthenticated write endpoint (see the OTLP note
# below).
listenAddress = "unix:${socketPath}";
};
};
};
@ -235,8 +345,8 @@ in
# deployment ever needs them.
#
# ⚠️ That endpoint is unauthenticated, which is why `listenAddress` above is
# loopback: the only route to it from off-host is the vhost's ingest
# location, which authenticates. The collector is still the only writer, but
# a socket only the gateway can reach: the only route to it is the vhost's
# ingest location, which authenticates. The collector is still the only writer, but
# it now arrives by the swarm name rather than over loopback, because it need
# not share a host with this store.
}

2
nix/module-eval/swarm-otel-identity.nix
View file

@ -282,7 +282,7 @@ let
m = locs."= /metrics" or null;
in
m != null
&& m.proxyPass == "http://127.0.0.1:${toString h.swarm.victoriametrics.port}/metrics"
&& m.proxyPass == "http://unix:/run/swarm-victoriametrics/victoriametrics.sock:/metrics"
&& lib.hasInfix "auth_request /__metrics_push_authz;" m.extraConfig
&& !(lib.hasInfix "error_page" m.extraConfig)
&& lib.hasInfix "internal;" locs."= /__metrics_push_authz".extraConfig

90
nix/module-eval/swarm-services-switch.nix
View file

@ -161,14 +161,31 @@ let
lib.filter (l: swarmServiceLinkDomains ? ${l.label}) cfg.services.hyperhive.swarm.controller.links;
# Every service container at once: the services-only host plus the CI
# runner. The CI runner and the review classifier are the two containers on
# a netns of their own.
# runner. The CI runner, the review classifier and the metrics store are the
# three containers on a netns of their own.
serviceContainersWithCi = hive {
deploy.hive-controller.enable = false;
deploy.allSwarmServices = true;
deploy.ci-runner.enable = true;
};
serviceContainers = serviceContainersWithCi.containers;
# The metrics store on a build older than its socket needs. Only its
# `assertions` are read, and the package is a stub carrying a version.
oldMetricsStore = hive {
deploy.victoriametrics.enable = true;
deploy.victoriametrics.package = pkgs.emptyDirectory // {
version = "1.152.0";
};
};
# Failing assertions first, then their messages: a message is only
# well-defined when its assertion fails (nixpkgs' fileSystems one throws
# otherwise).
failedMetricsStoreVersion =
host:
lib.filter (lib.hasInfix "services.hyperhive.deploy.victoriametrics.package must be") (
map (a: a.message) (lib.filter (a: !a.assertion) host.assertions)
);
cases = [
{
# `lib.all` over an empty set holds vacuously, so the roster is counted
@ -484,7 +501,7 @@ let
let
shared = lib.filter (c: !c.privateNetwork) (lib.attrValues serviceContainers);
in
lib.length shared == 9
lib.length shared == 8
&& lib.all (
c: !c.config.networking.firewall.enable && !c.config.networking.resolvconf.enable
) shared;
@ -499,6 +516,73 @@ let
in
c.networking.firewall.enable && c.networking.resolvconf.enable;
}
{
# No interface but loopback, so the socket is the store's only listener
# the host can reach. Every location that forwards to the store goes
# through it; the authelia subrequest is the one `https://` upstream.
# The read, push and scrape locations are named so an empty filter
# can't pass.
name = "the metrics store runs on a netns of its own and the gateway reaches it over its socket";
ok =
let
host = serviceContainersWithCi;
c = host.containers.swarm-victoriametrics;
dir = "/run/swarm-victoriametrics";
sock = "${dir}/victoriametrics.sock";
viaSock = path: "http://unix:${sock}:${path}";
locs =
host.services.nginx.virtualHosts.${host.services.hyperhive.swarm.victoriametrics.domain}.locations;
toStore = lib.filter (p: p != null && !(lib.hasPrefix "https://" p)) (
map (l: l.proxyPass) (lib.attrValues locs)
);
unit = c.config.systemd.services.victoriametrics;
nginxGid = host.ids.gids.nginx;
in
c.privateNetwork
&& c.hostBridge == null
&& c.hostAddress == null
&& c.config.services.victoriametrics.listenAddress == "unix:${sock}"
&& locs."= /api/v1/query".proxyPass == viaSock "/api/v1/query"
&& locs."= /opentelemetry/api/v1/push".proxyPass == viaSock "/opentelemetry/api/v1/push"
&& locs."= /metrics".proxyPass == viaSock "/metrics"
&& lib.all (lib.hasPrefix (viaSock "")) toStore
&& c.bindMounts.${dir}.hostPath == dir
&& builtins.elem "d ${dir} 0770 root ${toString nginxGid} - -" host.systemd.tmpfiles.rules
&& c.config.users.groups.gateway-nginx.gid == nginxGid
&& unit.serviceConfig.Group == "gateway-nginx"
&& !unit.serviceConfig.DynamicUser
&& unit.serviceConfig.UMask == "0007"
&& lib.hasInfix "--unix-socket ${sock}" unit.postStart;
}
{
# The DynamicUser layout keeps the history under /var/lib/private; the
# static user's StateDirectory is the plain path. The move runs as its
# own unit ahead of the store.
name = "the metrics store's history moves from the DynamicUser path before the store starts";
ok =
let
cfg = serviceContainers.swarm-victoriametrics.config;
m = cfg.systemd.services.victoriametrics-state-migrate;
in
cfg.services.victoriametrics.stateDir == "victoriametrics"
&& builtins.elem "victoriametrics.service" m.before
&& builtins.elem "victoriametrics.service" m.requiredBy
&& lib.hasInfix "old=/var/lib/private/victoriametrics" m.script
&& lib.hasInfix "new=/var/lib/victoriametrics" m.script
&& lib.hasInfix ''mv "$old" "$new"'' m.script
&& lib.hasInfix ''chown -R victoriametrics:gateway-nginx "$new"'' m.script;
}
{
# The control is the default package on the all-services host.
name = "a metrics store package older than 1.153.0 fails evaluation";
ok =
let
old = failedMetricsStoreVersion oldMetricsStore;
in
lib.length old == 1
&& lib.hasInfix "1.152.0" (lib.head old)
&& failedMetricsStoreVersion serviceContainersWithCi == [ ];
}
{
name = "a service container that sets no stateVersion of its own is on 26.05";
ok = serviceContainers.swarm-victoriametrics.config.system.stateVersion == "26.05";