Linux static binaries compiled 2026-10-17. Copy, chmod +x, run. Zero dependencies.
user line in a root-owned path refuses to start instead of silently running as root, and the CLI needs root or membership of group apm. Run sudo apm healconf after installing: it checks your config and every include, explains each setting this version reads differently, and fixes it with a backup beside the file (--check only reports; --yes applies every safe fix). apm upgrade runs the check for you.
Wacatac.B!ml. This is a known false positive: its ML heuristic misfires on statically compiled Go binaries and UPX-compressed executables. It has been reported to Microsoft for review. All other major antivirus engines report these files as clean — see the VirusTotal links below for full scan results.
metadata.json (the SHA-256 of every file above), install.sh and each connector carry an ed25519 signature beside them as <file>.sig. The public key is compiled into every APM binary since v2.10: apm upgrade refuses anything that does not verify, and apm verify <file> <file>.sig checks a download by hand. Key fingerprint (SHA-256 of the public key): f6af819d1b56bcc075a009f5051bb3a4bcf5d0cd962b513d763898a8ca09527f
-----BEGIN PUBLIC KEY----- MCowBQYDK2VwAyEAJaQi2HxEjH2ncrimkjGwyyj1XYYwfqTtSPuNT2VIiy8= -----END PUBLIC KEY-----To check with OpenSSL:
openssl pkeyutl -verify -pubin -inkey apm.pub -rawin -in metadata.json -sigfile <(base64 -d metadata.json.sig)
Until now a daemon told the outside world one thing: the anonymous hourly telemetry ping, which knows no names and draws no graphs. Everything else — is the API up, why did the cron worker die at 03:12, what did it print before it died — meant ssh and apm log. This release adds the daemon side of the Portal, the hosted dashboard at portal.processmanager.dev for every daemon an account runs: a reporter, a live link, and remote control that stays off until the operator turns it on. Nothing is sent without a key, and telemetry is unchanged either way.
portal_key — the daemon reports (opt-in)portal_key apmk_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX; # top level: every worker reports under this key
portal_name "prod-eu-1"; # optional, default: hostname
portal_interval 30; # optional, seconds, 10–600 (the portal may raise it)
worker {
name api;
portal_key apmk_YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY; # this worker reports under its own key
}
With a key, the daemon reports every portal_interval seconds: itself (installation id, name, version, the machine's CPU %, memory and load) and each of that key's workers — status, instances, CPU and RSS as average and peak, connections, restarts, children — plus the lifecycle events since the last report, child_exit with its exit code. A crash or a reload brings the next report forward to a few seconds; systemctl stop apm sends a last one. Keys scope by file: a worker's own portal_key, else the top-level key of the conf file that declares it, else apm.conf's — so apm.conf.d/shop.conf can report its workers to another account while the rest report to the main one. Keys are re-read every 5 s: apm reload starts, moves or stops reporting without restarting anything. A rejected key is logged once (keys appear only as … + their last 4 characters) and retried every 10 minutes; outages back off from 30 s to 10 minutes and log only the first failure and the recovery. apm portal shows every key, its workers, the last result and when the next report goes.
Each key also keeps one WebSocket to the portal; reports travel over it, and over HTTPS POST while it is down. While someone watches the daemon, the link carries live figures every few seconds — current CPU, RSS and connections per worker — and stops by itself 10 minutes after the last request. Where the package has them, each worker also reports its advanced stats: Vanguard rejections, bans and blocked connections (new per-worker counters, kept across a rule reload), micro-dedupe hits, requests, its children's disk I/O and open file descriptors, WebSocket clients and the 5-minute ping round trip. Live figures are never stored.
portal_control — remote control and logs, off by defaultportal_control true; # top level, default off
Only with portal_control true; AND a package that allows it, the portal can restart, stop, start or reload a worker; list, read and write the conf files the key owns — a write is refused when the file changed on disk since it was read, when it does not parse, or when it touches an include line; the old file is kept as <file>.bak-<unix time> (which include apm.conf.d/*; and apm healconf skip) and applying touches only the workers that file declares; and read a worker's log — the last 500 lines of its log or err_log, read the way apm log reads them, plus a live follow of its children's stdout and stderr while someone has the log open. A key reaches only its own workers and files: a file's own key never sees apm.conf. One command runs at a time and every one is in the daemon log: portal: cmd worker.restart api (key …abcd): ok. A plain http:// portal_url is accepted for loopback only.
Content-Encoding (gzip, br,deflate) now reaches the worker as bytes whatever its Content-Type says; before, a compressed JSON body was handed over as text and mangled.
apm load of a file with its own top-level portal_key no longer lets anearly report of the new worker reach the daemon-wide key.
apm healconf counts portal_key as a secret in a world-readable confand skips .bak-<unix time> copies, saying how many it skipped.
Upgrade notes: nothing changes unless you set portal_key. portal_control is off by default — turn it on only on daemons whose keys you trust with restarts, conf files and logs. Details in the timestamped notes beside this file.