Repository navigation
Lock
A named lock only one process can hold at a time, across every host that shares the database.
use Blink\Lock;
Lock::of("reports:rebuild") -> run(function () {
Reports::rebuild();
}, ttl: 600, wait: 30);run() acquires, runs the callable, releases, and rethrows anything the callable threw. wait is how long to keep trying before giving up with LockAcquired; 0 tries once.
By hand:
$lock = Lock::of("imports:daily");
if (!$lock -> acquire(ttl: 300))
return;
try {
Importer::run();
$lock -> refresh(300); // a long job extends its own claim
} finally {
$lock -> release();
}| Method | Does |
|---|---|
acquire($ttl, $throw) |
Take the lock now, or return false. Takes over an expired claim. |
block($wait, $ttl, $throw) |
Retry for up to $wait seconds |
release() |
Give it back, if this instance still holds it |
refresh($ttl) |
Push the expiry out |
held() |
Whether anyone holds it |
holder() |
Who: host, pid, when, and where in the code |
Every lock expires. A holder that crashes leaves its claim behind, and the expiry is what lets the next process continue instead of waiting on one that is never coming back. It also means a holder that overruns its TTL no longer owns the lock; anything it does after that point has to be safe to happen twice, or it should refresh() along the way. The default TTL is Lock::DEFAULT_TTL.
Lock::of() uses CONFIG::$LOCK_STORE, database by default: a row in the locks table, which works across hosts. Lock::file() uses a file under DATA_ROOT, for the one lock that has to exist before the table does, Terra's own. Lock::database() names the database store explicitly.
php core/blink lock:list
php core/blink lock:release reports:rebuild --force
lock:list shows every held lock with its holder, which is where an orbit entry marked alone() shows up while it runs.