feat: migrate to resticontainer with label-driven per-service backup

Swap the restic core service for resticontainer
(ghcr.io/deytenit/resticontainer:0.19.1): per-service backup intent via
restic.* compose labels (paths / stop / pre+post hooks), docker socket +
/:/hostfs:ro to resolve container mounts to host paths and snapshot the
union in one restic run. backup/run drops explicit /data/tier1,2 args
(paths now come from labels), keeps init/forget/notify, and guards on a
fresh snapshot. Snapshots are stamped --host <node>.apex.ermnvldmr.com
--tag biweekly and written with --compression max.

PARITY.md

@@ -14,7 +14,7 @@ action to its bash source and the behaviors that must match.
 | configure | (new umbrella) | run base,ufw,crowdsec,cron,systemd,routing in order |
 | sync/repository | actions/sync/repository | sync/<node> branch; add -A + commons submodule; commit; rebase origin/main (was trunk in the monorepo); force-with-lease; telegram |
 | sync/packages | actions/sync/packages | apt --just-print upgrade Inst list; docker+skopeo digest compare; telegram info |
-| backup/run | actions/sync/tiers (restic-backup) | init-if-needed; backup /data/@tier1,2 --host --tag com-ermnvldmr-root-<node>; forget --prune 7/4/12; telegram |
+| backup/run | actions/sync/tiers (restic-backup) | init-if-needed; label-driven backup via resticontainer (discovers `restic.*`-labelled services, runs pre-hooks/stops, snapshots the resolved host paths) --host <node>.apex.ermnvldmr.com --tag biweekly --compression max; guard on a fresh snapshot; forget --prune 7/4/12; telegram |
 | tiers/link | actions/tiers/link | node + per-project @tierN symlinks; shared/ links; .env symlink |
 | tiers/useradd | actions/tiers/useradd | noroot-<proj> system users; noroot-shared group; APEX_UID/GID block in .env |
 | tiers/chown | actions/tiers/chown | {tier}/{proj}=uid:gid; {tier}/shared/{proj}=uid:sgid+sticky; base=root:root |
@@ -26,7 +26,8 @@ action to its bash source and the behaviors that must match.
 ## Deliberate deltas (rename + fixes)
 - root->apex everywhere (CLI, container names apex-*, telegram instance com.ermnvldmr.apex.*, ufw markers APEX.*, networks direct/enclave/socket).
 - No <node> arg; identity from FQDN + node.env.
-- restic core service gains read-only /data/@tier1,2 mounts (makes existing backup command functional).
+- backup superseded by the resticontainer migration (see below): the old whole-tier-root
+  restic snapshot is replaced by per-service, `restic.*`-label-driven backups.
 - ufw markers self-heal (delete ROOT + APEX blocks before inject).
 - envsubst -> string.Template.safe_substitute; wget/curl download -> urllib. Unset template
   vars stay LITERAL (envsubst emptied them) — typos surface instead of silently blanking.
@@ -57,3 +58,18 @@ action to its bash source and the behaviors that must match.
 - restic: per-node selective tier2 mounts replaced by whole-tier-root RO mounts
   (superset; snapshot paths preserved); RESTIC_COMPRESSION per-node via
   `APEX_RESTIC_COMPRESSION` (default auto).
+
+## resticontainer migration (SP4 realized)
+- The `restic` core service is replaced by `resticontainer`
+  (`ghcr.io/deytenit/resticontainer`), a restic wrapper that reads per-service
+  `restic.*` compose labels: `restic.enable`, `restic.backup.paths` (comma-separated
+  container mount destinations — matched exactly, whole mount dir backed up),
+  `restic.backup.stop`, and `restic.hooks.pre-backup`/`post-backup` (run via docker exec).
+- It mounts the docker socket (list/inspect/exec/stop/start) and the host root
+  `/:/hostfs:ro` (`RESTIC_HOSTFS=/hostfs`); it resolves each labelled mount to
+  `/hostfs` + the mount's host source and snapshots the union in one restic run.
+- The whole-tier-root RO mounts and the explicit `/data/@tier1,2` backup args are gone;
+  a service is backed up only if it carries `restic.enable=true`. DB consistency is
+  per-service: a pre-backup hook dumps into a dedicated backed-up mount (e.g. postgres
+  `pg_dump … > /dumps/dump.sql`), never the live data dir. `backup/run` now guards on a
+  fresh snapshot landing for the host (a no-label repo would otherwise be a silent no-op).

README.md

@@ -3,6 +3,8 @@
 The **commons** of the apex fleet: a stock-python engine (`apex <group>/<action>`) and
 the shared core docker-compose, consumed by every node as the `commons/` git submodule.
 
+Full documentation: [docs.ermnvldmr.com/en/apex](https://docs.ermnvldmr.com/en/apex/).
+
 ## Layout
 - `apex` — launcher (python3, stdlib only); node `init.sh` symlinks it onto `PATH`.
 - `engine/` — the runner: CLI, identity, overlay, context, and the `ctx.*` helper library.

actions/backup/run.py

@@ -1,17 +1,23 @@
 # actions/backup/run.py
-"""Port of sync/tiers restic-backup (SP4: behavior-identical). Runs the restic core
-service one-shot: init-if-needed -> backup tiers -> forget --prune -> Telegram."""
+"""Label-driven restic backup via resticontainer (github.com/deytenit/resticontainer).
+Runs the resticontainer core service one-shot: init-if-needed -> backup (discovers
+`restic.*`-labelled services, runs their pre-hooks/stops, snapshots the union of the
+resolved host paths in a single restic run) -> forget --prune -> Telegram. Every
+non-`backup` subcommand passes straight through to restic, so init/snapshots/forget
+behave exactly as before."""
+import json
 import os
 from engine.descriptor import Meta, Arg
 
-METADATA = Meta(summary="Run restic backup of the backed-up tiers (init-if-needed, backup, forget --prune).",
+METADATA = Meta(summary="Run the resticontainer label-driven backup (init-if-needed, backup, forget --prune).",
                 args=[Arg("telegram_bot_url", "Telegram Bot URL for notifications")])
 TITLE = "Restic Host Backup"
 
 def run(ctx, args):
     log, node, url = ctx.log, ctx.node.name, args.telegram_bot_url
     core = ctx.paths.core
-    tag = f"com-ermnvldmr-root-{node}"          # KEPT unchanged for backup continuity (SP4/Decision D)
+    host = f"{node}.apex.ermnvldmr.com"         # restic --host: the per-node public name (matches the bucket)
+    tag = "biweekly"                            # restic --tag: snapshot cadence (this cron fires on the 5th & 20th)
     if not os.path.isfile(os.path.join(core, "docker-compose.yml")):
         log.error(f"Core compose not found in {core}"); raise SystemExit(1)
     env = dict(os.environ, COMPOSE_PROFILES="manual")
@@ -21,9 +27,20 @@ def run(ctx, args):
         if os.path.isfile(os.path.join(core, ef)):
             envfiles += ["--env-file", ef]
 
-    def restic(*cmd, check=True):
+    def restic(*cmd, check=True, capture=False):
         return ctx.sys.run(["docker", "compose", *envfiles, "--profile", "manual",
-                            "run", "--rm", "restic", *cmd], cwd=core, env=env, check=check)
+                            "run", "--rm", "-T", "resticontainer", *cmd],
+                           cwd=core, env=env, check=check, capture=capture)
+
+    def host_snapshot_ids():
+        """This host's snapshot IDs, or None if it can't be determined (never fail on this)."""
+        cp = restic("snapshots", "--host", host, "--json", check=False, capture=True)
+        if cp.returncode != 0:
+            return None
+        try:
+            return {s.get("id") for s in json.loads(cp.stdout or "[]")}
+        except (ValueError, TypeError):
+            return None
 
     try:
         log.info("Checking restic repository availability...")
@@ -31,10 +48,20 @@ def run(ctx, args):
             log.info("Restic repository reachable.")
         elif restic("init", check=False).returncode != 0:
             ctx.notify.error(TITLE, url, "restic init failed", node); raise SystemExit(1)
-        log.info("Backing up /data/@tier1 and /data/@tier2 ...")
-        if restic("backup", "/data/@tier1", "/data/@tier2", "--host", node, "--tag", tag,
-                  "--compression", "auto", check=False).returncode != 0:
+        before = host_snapshot_ids()
+        log.info("Backing up `restic.*`-labelled services (resticontainer discovers paths)...")
+        if restic("backup", "--host", host, "--tag", tag, "--compression", "max",
+                  check=False).returncode != 0:
             ctx.notify.error(TITLE, url, "restic backup failed", node); raise SystemExit(1)
+        # Guard: the label-driven backup is a silent no-op if no service carries the
+        # `restic.enable` label (resticontainer finds zero paths and skips restic).
+        # Prove a fresh snapshot for this host actually landed before we prune/notify.
+        after = host_snapshot_ids()
+        if before is not None and after is not None and not (after - before):
+            log.error("No new snapshot produced — refusing to report success.")
+            ctx.notify.error(TITLE, url,
+                             "backup produced no new snapshot (no restic.* labels discovered?)", node)
+            raise SystemExit(1)
         log.info("Applying retention (forget --prune)...")
         if restic("forget", "--prune", "--keep-daily", "7", "--keep-weekly", "4",
                   "--keep-monthly", "12", check=False).returncode != 0:

actions/tiers/chown.py

@@ -46,6 +46,16 @@ def run(ctx, args):
         full = os.path.join(comp, proj)
         if not os.path.isdir(full) or proj.startswith(".") or proj.startswith("@"):
             continue
+        # opt-out: .apex-root-owned forces this project's tier data (dir + children,
+        # recursively) to root:root — for services that run as root and reject
+        # non-root-owned data (e.g. seafile-mc). Blocks noroot ownership entirely.
+        if os.path.isfile(os.path.join(full, ".apex-root-owned")):
+            log.info(f"Applying root ownership for {proj} (.apex-root-owned)")
+            for tp in tiers:
+                for d in (os.path.join(tp, proj), os.path.join(tp, "shared", proj)):
+                    if os.path.isdir(d):
+                        s.sudo(["chown", "-R", "root:root", d]); log.success(f"  Owned {d} (root)")
+            continue
         env_path = os.path.join(full, ".env")
         uid = _env_val(env_path, "APEX_UID")
         gid = _env_val(env_path, "APEX_GID")

actions/tiers/useradd.py

@@ -34,6 +34,12 @@ def run(ctx, args):
         full = os.path.join(comp, proj)
         if not os.path.isdir(full) or proj.startswith(".") or proj.startswith("@"):
             continue
+        # opt-out: a project marked .apex-root-owned runs as root and keeps its tier
+        # data root-owned (e.g. seafile-mc, whose entrypoint refuses non-root-owned
+        # data). Skip the noroot user + .env UID injection; tiers/chown forces root.
+        if os.path.isfile(os.path.join(full, ".apex-root-owned")):
+            log.info(f"{proj}: .apex-root-owned — runs as root; skipping noroot user")
+            continue
         base = f"noroot-{_sanitize(proj)}"
         cand, i = base, 1
         while cand in used:

assets/alloy/config.alloy

@@ -0,0 +1,120 @@
+// ============================================================================
+// APEX fleet observability agent — per-node Grafana Alloy config (shared).
+// Committed, env-parameterized: every node runs this SAME file; per-node values
+// come from sys.env(...) (APEX_NODE, push URLs, obs basic-auth creds).
+// Verified against grafana.com/docs/alloy/latest (Alloy v1.17.1).
+// ============================================================================
+// Model: PUSH. Metrics -> remote_write to the central Prometheus (on icarus);
+// logs -> loki.write to the central Loki (on icarus). daedalus/morpheus reach
+// them through the public gfn.ermnvldmr.com front (authelia-bypassed + basicauth
+// on /loki and /prometheus/api/v1/write); icarus resolves gfn locally.
+// ============================================================================
+
+// ----------------------------------------------------------------------------
+// METRICS
+// ----------------------------------------------------------------------------
+
+// Host metrics (node_exporter-equivalent). Container mounts host /proc,/sys,/
+// at /host/proc,/host/sys,/rootfs so the exporter reads the HOST.
+prometheus.exporter.unix "node" {
+  procfs_path = "/host/proc"
+  sysfs_path  = "/host/sys"
+  rootfs_path = "/rootfs"
+}
+
+// Container metrics (cadvisor-equivalent) via the docker socket.
+prometheus.exporter.cadvisor "containers" {
+  docker_host            = "unix:///var/run/docker.sock"
+  store_container_labels = true
+}
+
+// Stamp a static `node` label onto each exporter's targets.
+discovery.relabel "node_metrics" {
+  targets = prometheus.exporter.unix.node.targets
+  rule {
+    action       = "replace"
+    target_label = "node"
+    replacement  = sys.env("APEX_NODE")
+  }
+}
+discovery.relabel "container_metrics" {
+  targets = prometheus.exporter.cadvisor.containers.targets
+  rule {
+    action       = "replace"
+    target_label = "node"
+    replacement  = sys.env("APEX_NODE")
+  }
+}
+
+prometheus.scrape "node" {
+  targets         = discovery.relabel.node_metrics.output
+  forward_to      = [prometheus.remote_write.default.receiver]
+  scrape_interval = "15s"
+}
+prometheus.scrape "containers" {
+  targets         = discovery.relabel.container_metrics.output
+  forward_to      = [prometheus.remote_write.default.receiver]
+  scrape_interval = "15s"
+}
+
+// Push metrics to the central Prometheus (needs --web.enable-remote-write-receiver).
+prometheus.remote_write "default" {
+  endpoint {
+    url = sys.env("APEX_PROM_PUSH_URL")
+    basic_auth {
+      username = sys.env("APEX_OBS_BASIC_AUTH_USER")
+      password = sys.env("APEX_OBS_BASIC_AUTH_PASS")
+    }
+  }
+}
+
+// ----------------------------------------------------------------------------
+// LOGS
+// ----------------------------------------------------------------------------
+
+// systemd journal (host services, ufw, sshd, crowdsec, etc.).
+loki.source.journal "systemd" {
+  path    = "/var/log/journal"
+  max_age = "12h"
+  labels  = {
+    node = sys.env("APEX_NODE"),
+    job  = "systemd-journal",
+  }
+  forward_to = [loki.write.default.receiver]
+}
+
+// All docker containers' stdout/stderr.
+discovery.docker "containers" {
+  host = "unix:///var/run/docker.sock"
+}
+discovery.relabel "container_logs" {
+  targets = discovery.docker.containers.targets
+  rule {
+    action        = "replace"
+    source_labels = ["__meta_docker_container_name"]
+    regex         = "/?(.*)"
+    target_label  = "container"
+  }
+  rule {
+    action       = "replace"
+    target_label = "node"
+    replacement  = sys.env("APEX_NODE")
+  }
+}
+loki.source.docker "containers" {
+  host          = "unix:///var/run/docker.sock"
+  targets       = discovery.relabel.container_logs.output
+  relabel_rules = discovery.relabel.container_logs.rules
+  forward_to    = [loki.write.default.receiver]
+}
+
+// Push logs to the central Loki.
+loki.write "default" {
+  endpoint {
+    url = sys.env("APEX_LOKI_PUSH_URL")
+    basic_auth {
+      username = sys.env("APEX_OBS_BASIC_AUTH_USER")
+      password = sys.env("APEX_OBS_BASIC_AUTH_PASS")
+    }
+  }
+}

compositions/apex/docker-compose.yml

@@ -27,30 +27,39 @@ services:
       - no-new-privileges:true
 
   # --- backups (host network, manual profile, one-shot) ---
-  restic:
-    image: restic/restic:0.18.1
-    container_name: apex-restic
+  # resticontainer (github.com/deytenit/resticontainer): a restic wrapper that
+  # discovers per-service backup intent from `restic.*` compose labels
+  # (restic.enable / restic.backup.paths / restic.backup.stop / restic.hooks.*).
+  # On `backup` it queries the docker daemon, runs each service's pre-hook (e.g.
+  # pg_dump), stops any container flagged restic.backup.stop, resolves the labeled
+  # container mounts to host paths (via /hostfs + mount source), snapshots the union
+  # in a single restic run, then restarts + runs post-hooks. Every other subcommand
+  # (init/snapshots/forget/…) passes straight through to restic. Needs the docker
+  # socket (list/inspect/exec/stop/start) and the host root read-only (to read the
+  # data) — a privileged manual one-shot, same trust surface as the old restic core.
+  resticontainer:
+    image: ghcr.io/deytenit/resticontainer:0.19.1
+    container_name: apex-resticontainer
     profiles: [manual]
     volumes:
       - ${APEX_TIER3}/restic/cache:/var/restic-cache
-      # FIX (Task 1.11): mount the backed-up tier roots read-only so `restic backup
-      # /data/@tier1 /data/@tier2` actually snapshots data (was missing today).
-      - ${APEX_TIER_ROOT1}:/data/@tier1:ro
-      - ${APEX_TIER_ROOT2}:/data/@tier2:ro
+      - /var/run/docker.sock:/var/run/docker.sock          # query + exec/stop/start
+      - /:/hostfs:ro                                        # read the data off the host
     environment:
       <<: *shared-environment
+      RESTIC_HOSTFS: /hostfs                                # prefix for resolved mount sources
       RESTIC_REPOSITORY: "${APEX_RESTIC_REPOSITORY?error}"
       RESTIC_PASSWORD: "${APEX_RESTIC_PASSWORD?error}"
       AWS_ACCESS_KEY_ID: "${APEX_RESTIC_AWS_ACCESS_KEY_ID?error}"
       AWS_SECRET_ACCESS_KEY: "${APEX_RESTIC_AWS_SECRET_ACCESS_KEY?error}"
       AWS_REGION: "ru-central1"
       AWS_ENDPOINT_URL: "https://storage.yandexcloud.net/"
-      # per-node override (icarus historically ran "max" for manual invocations;
-      # scheduled backups always force --compression auto on the CLI)
-      RESTIC_COMPRESSION: "${APEX_RESTIC_COMPRESSION:-auto}"
+      # max compression is fleet policy (backup/run also forces --compression max on
+      # the CLI; needs repo format v2, which a fresh restic init creates). Per-node
+      # override via APEX_RESTIC_COMPRESSION only if ever needed.
+      RESTIC_COMPRESSION: "${APEX_RESTIC_COMPRESSION:-max}"
       RESTIC_CACHE_DIR: /var/restic-cache
     network_mode: "host"
-    entrypoint: ["restic"]
 
   # --- docker socket proxy ---
   socket-proxy:
@@ -152,27 +161,48 @@ services:
     security_opt:
       - no-new-privileges:true
 
-  # --- optional: loki log aggregation (node-local, enclave) ---
-  loki:
-    image: grafana/loki:3.1.1
-    container_name: apex-loki
-    profiles: [loki]
-    command: ["-config.file=/etc/loki/config.yaml"]
+  # --- optional: observability agent (profile-gated) ---
+  # Grafana Alloy: one agent per node. Collects host metrics (node_exporter-
+  # equivalent) + container metrics (cadvisor) and PUSHES them via remote_write;
+  # collects systemd-journal + docker logs and PUSHES them to Loki. All targets
+  # are OUTBOUND to the central gfn stack (Prometheus + Loki live on icarus), so
+  # Alloy needs egress (direct, not enclave). Replaces the old store-only `loki`
+  # service (which had no feeder). Config is a committed, env-parameterized asset.
+  alloy:
+    image: grafana/alloy:v1.17.1
+    container_name: apex-alloy
+    profiles: [alloy]
+    user: "0:0"  # root: required for host metrics, journal, and docker.sock
     volumes:
-      - ${APEX_TIER1}/loki/config.yaml:/etc/loki/config.yaml:ro
-      - ${APEX_TIER2}/loki/data:/loki
+      - ${APEX_COMMONS}/assets/alloy/config.alloy:/etc/alloy/config.alloy:ro
+      - ${APEX_TIER2}/alloy/data:/var/lib/alloy
+      - /var/run/docker.sock:/var/run/docker.sock:ro
+      - /proc:/host/proc:ro
+      - /sys:/host/sys:ro
+      - /:/rootfs:ro
+      - /var/log/journal:/var/log/journal:ro
+      - /run/log/journal:/run/log/journal:ro
+      - /etc/machine-id:/etc/machine-id:ro
+    environment:
+      <<: *shared-environment
+      APEX_NODE: "${APEX_NODE?error}"
+      # push targets + machine auth (per-node scalars in apex.env / secret .env)
+      APEX_LOKI_PUSH_URL: "${APEX_LOKI_PUSH_URL?error}"
+      APEX_PROM_PUSH_URL: "${APEX_PROM_PUSH_URL?error}"
+      APEX_OBS_BASIC_AUTH_USER: "${APEX_OBS_BASIC_AUTH_USER?error}"
+      APEX_OBS_BASIC_AUTH_PASS: "${APEX_OBS_BASIC_AUTH_PASS?error}"
+    command:
+      - run
+      - --server.http.listen-addr=0.0.0.0:12345
+      - --storage.path=/var/lib/alloy
+      - /etc/alloy/config.alloy
     networks:
-      enclave: {}
+      direct: {}
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     restart: unless-stopped
-    user: "${APEX_UID?error}:${APEX_GID?error}"
-    labels:
-      # /loki/* is loki's native API prefix — no strip needed. basicauth users come
-      # from the node's uncommitted .env (only needed when the loki profile is on).
-      - "traefik.enable=true"
-      - "traefik.http.routers.apex-loki.rule=Host(`${APEX_NODE_FQDN}`) && PathPrefix(`/loki`)"
-      - "traefik.http.routers.apex-loki.middlewares=apex-loki-auth"
-      - "traefik.http.middlewares.apex-loki-auth.basicauth.users=${APEX_LOKI_BASIC_AUTH}"
-      - "traefik.http.services.apex-loki.loadbalancer.server.port=3100"
+    security_opt:
+      - no-new-privileges:true
 
 networks:
   direct: