elastic_fleet_load_integrations_dir now buffers each concurrent job's
output (header + API response) to a per-job file and prints them in
submission order after wait, restoring the readable serial-style output
while keeping concurrent writes.
Add --retry-all-errors to the integration create/update curl calls so
transient 409 conflicts from concurrent writes to the same agent policy
are retried (curl --retry alone does not retry 409).
Fetch each agent policy once and extract integration name/package/version/id
locally via a single jq pass instead of re-fetching the identical policy JSON
1+3N times. Memoize epm/packages latest-version lookups so each package is
queried once instead of per (policy, integration). Dispatch the per-integration
dry-run+upgrade as throttled background jobs (MAX_FLEET_JOBS) with
flock-serialized output and a FAIL_FILE marker, mirroring
elastic_fleet_load_integrations_dir.
Behavior preserved: same elastic-defend-endpoints/fleet_server skips, same
AUTO_UPGRADE_INTEGRATIONS default-package gating (moved into jq, using $defaults
to avoid the jq $def keyword collision), and exit 1 on any failure so salt
retries.
Load component and index templates as throttled background jobs (max 10
concurrent) instead of sequential curl PUTs, matching the bounded-concurrency
+ flock-serialized-output pattern used by the fleet/ILM load scripts. Keeps a
wait barrier between the component phase and the index phase so index
templates never load before their referenced component templates. Failures are
tracked via per-job marker files since counter increments can't escape
background subshells.
Fetch each agent policy once per group instead of refetching the full
policy (plus a fresh Kibana session cookie) for every integration file,
and dispatch the create/update writes as throttled background jobs.
Adds elastic_fleet_load_integrations_dir and elastic_fleet_throttle to
so-elastic-fleet-common, reusing the bounded-concurrency pattern from
so-elasticsearch-ilm-policy-load. Replaces the four serial loops in the
loader with one call per agent policy.
The agent-policy enumeration passed --argjson def, creating a jq
variable $def. 'def' is a reserved keyword in jq and the deployed jq
version rejects it, so the program failed to compile and
in_use_integrations was left empty (silently disabling the in-use
upgrade guard). Rename the arg to $defaults.
A single printf per block was not actually one write() call, so
concurrent jobs still occasionally interleaved their label and response
lines. Hold an flock around just the printf (curl still runs in
parallel) so each policy's block prints intact, keeping live
completion-order streaming.
Replace the per-package decision loop (which forked ~10 processes per
package and rebuilt a growing JSON file on every add -> O(n^2)) with two
jq passes: one prints the status messages, one builds the bulk install
list. A vnum/needs() jq definition reproduces the previous
version_conversion/compare_versions and excluded/subscription/installed/
upgrade/in-use logic exactly. Also fetch each agent policy once and
extract non-default package names locally instead of re-fetching the
policy per integration (1+K -> 1 GET per policy). Install behavior is
unchanged.
Run the ~300 ILM policy PUTs concurrently (bounded to 10 in flight via a
throttle gate) instead of one serial curl per policy. Adds a put_policy
helper and waits for all background jobs before exiting. Preserves policy
parity; only the scheduling changes. Drops the dead empty sid cookie arg
(falls back to basic auth from curl.config as before).
A complete mine is not enough: elasticsearch:nodes, redis:nodes,
logstash:nodes (tgt_type=pillar) and hypervisor:nodes (tgt_type=compound)
resolve their target against the master's per-minion data cache
(grains+pillar in data.p), which is populated only when a minion's pillar
is recompiled -- separately from the mine. After a reboot a node can be in
the mine (so node_data/glob sees it) yet absent from that cache, so it
fails the elasticsearch:enabled:true pillar match and is dropped from
elasticsearch:nodes -> so-elasticsearch ExtraHosts -> container recreate.
After the mine-completeness wait, run salt '*' saltutil.refresh_pillar
wait=True to synchronously cache every up node's pillar (the same lever
deploy_newnode.sls uses), then verify with salt-run cache.pillar and retry
stragglers, bounded by MINE_UPDATE_MAX_WAIT. Also log elasticsearch:nodes
alongside node_data for inspection.
Mine-backed pillars (node_data, elasticsearch:nodes, redis:nodes,
logstash:nodes, hypervisor:nodes) include a node only if it returned an
IP from the mine, and the configs they build are rebuilt fresh every
highstate. After a manager reboot with a flushed mine, the first boot
highstate could run before an up node re-reported network.ip_addrs,
dropping it from e.g. so-elasticsearch ExtraHosts and forcing a
container recreate.
After the initial broad mine.update, poll until every currently-up
minion actually has network.ip_addrs in the mine, re-pushing mine.update
to stragglers, before releasing the boot highstate. Shares the existing
MINE_UPDATE_MAX_WAIT backstop so a slow/down node never blocks boot, and
still logs the rendered node_data for inspection.
Dump the actual rendered node_data pillar (pretty-printed JSON) to the
journal instead of just a rendered/empty verdict, so the boot-time render
attempt is fully inspectable. Empty renders print false/null and still
emit the WARNING.
After the boot-time mine.update, have the manager actually render the
node_data pillar and log whether it came back populated. node_data: False
makes salt/top.sls apply the bootstrap recovery branch instead of the
manager's real config, so surfacing this in the journal makes the
condition visible before so-boot-highstate runs. Best-effort and
non-blocking: always exits 0 so highstate proceeds regardless.
so-boot-mine-update.service is a manager-only Type=oneshot unit that runs
once per boot after salt-master/salt-minion start and before
so-boot-highstate.service. It pushes mine.update to all reachable minions
so mine-backed pillars (node IPs, ES/Redis/Logstash discovery) are fresh
before the boot highstate renders them.
The helper waits for the responsive minion set to settle (plateau) rather
than for every accepted key to report up, so an intentionally powered-off
minion doesn't block the update; MAX_WAIT remains as a backstop.
The setup-complete marker is a runtime-state file, not config, so move it
to /opt/so/state/setup-complete. Updates both writers (mark_setup_complete
in setup/so-functions and the upgrade-path state in minion/init.sls) and the
three readers (so-boot-highstate.service ConditionPathExists, boot_highstate.sls
enable gate, and the so-user_sync cron gate).
Add docker.service to After= and Wants= so the boot-time highstate
starts after docker is up. Uses Wants (soft) so highstate still runs
if docker fails to start.
Josh Patterson
reyesj2
Jason Ertel
Jason Ertel
Josh Brower