Distributed* locking using PostgreSQL advisory locks.
Some use cases:
-
You have a clock process and want to make absolutely sure there will never be more than one process active at any given time.
This sort of situation can otherwise arise if the clock process is scaled up by accident or during a deployment which keeps the old version running until the new version responds to a health check.
-
Running a database migration at server startup. If your app is scaled, multiple processes will simultaneously try to run the database migration which can lead to problems.
-
Leader election. Let's say you have a web app and want to post a message to Slack every 30 mins containing some statistic (e.g. new registrations in the last 30 mins). You might have 10 processes running but don't want to get 10 identical messages in Slack. You can use this library to elect a "master" process which is responsible for sending the message.
* Your PostgreSQL database being a central point of failure. For a high available distributed lock, have a look at ZooKeeper.
npm install --save advisory-lock
A withlock
command line utility is provided to make to facilitate the
common use case of ensuring only one instance of a process is running at any
time.
withlock <lockName> [--db <connectionString>] -- <command>
Where <lockName>
is the name of the lock, <command>
(everything after
--
) is the command to run exclusively, once the lock is acquired.
--db <connectionString>
is optional and if not specified, the
PG_CONNECTION_STRING
environment variable will be used.
Example:
export PG_CONNECTION_STRING="postgres://postgres@127.0.0.1/mydb"
withlock dbmigration -- npm run knex migrate:latest
connectionString
must be a Postgres connection string
Returns a createMutex
function.
The createMutex
function also exposes a client
property
that can be used to terminate the database connection if necessary.
PS: Each call to advisoryLock(connectionString)
creates a new PostgreSQL
connection which is not automatically terminated, so if that is an
issue for you, you
can use createMutex.client.end()
to end the connection when
appropriate (e.g. after releasing a lock). This is however typically
not needed since usually, advisoryLock()
only needs to be called once.
lockName
must be a unique identifier for the lock
Returns a mutex object containing the functions listed below. All
object methods are really just functions attached to the object and
are not bound to this so they can be safely destructured,
e.g. const { withLock } = createMutext(lockName)
.
For a better understanding of what each functions does, see PosgtreSQL's manual.
fn
Promise returning function or regular function to be executed once the lock is acquired
Like lock()
but automatically release the lock after fn()
resolves.
Returns a promise which resolves to the value fn
resolves to.
Throws an error if the Postgres connection closes unexpectedly.
Returns a promise which resolves to true
if the lock is free and
false
if the lock is taken. Doesn't "block".
Wait until we get exclusive lock.
Release the exclusive lock.
Like tryLock()
but for shared lock.
While held, this blocks any attempt to obtain an exclusive lock. (e.g.: calls to .lock()
or .withLock()
)
Release shared lock.
Same as withLock()
but using a shared lock.
import advisoryLock from 'advisory-lock'
const mutex = advisoryLock('postgres://user:pass@localhost:3475/dbname')('some-lock-name')
const doSomething = () => {
// doSomething
return Promise.resolve()
}
mutex
.withLock(doSomething) // "blocks" until lock is free
.catch((err) => {
// this gets executed if the postgres connection closes unexpectedly, etc.
})
.then(() => {
// lock is released now...
})
// doesn't "block"
mutex.tryLock().then((obtainedLock) => {
if (obtainedLock) {
return doSomething().then(() => mutex.unlock())
} else {
throw new Error('failed to obtain lock')
}
})
See ./test for more usage examples.