class Prostore::Migration::HeartbeatFiber

Overview

Background fiber that renews the migration lease while a long step is running (ADR-0009 invariant 5).

Without a heartbeat, the lease only renews between steps in the runner's loop. A single multi-minute step (large eager-Crystal-lambda backfill, slow CREATE INDEX CONCURRENTLY, big table rebuild on SQLite) can outlast the lease and lose it to another runner.

The fiber wakes every interval (default 1 minute), tries to claim a connection from the pool, calls Lease.heartbeat. If the pool can't spare a connection (e.g., in-memory SQLite with pool size 1) the heartbeat is silently skipped — the runner's between-steps heartbeat is still in play, which is sufficient for the SQLite case where steps are fast anyway. If Lease.heartbeat reports the lease was stolen, the fiber sets #lost? and the runner aborts before the next step.

Cleanly stoppable via a Channel; #stop waits for the fiber to actually exit so the connection can be closed safely.

Defined in:

prostore/migration/heartbeat.cr

Constant Summary

DEFAULT_INTERVAL = 1.minute

Constructors

.new(conn : Connection, migration_id : Int64, runner_id : String, interval : Time::Span = DEFAULT_INTERVAL, lease_duration : Time::Span = Lease::DEFAULT_DURATION)

Constructor Detail

def self.new(conn : Connection, migration_id : Int64, runner_id : String, interval : Time::Span = DEFAULT_INTERVAL, lease_duration : Time::Span = Lease::DEFAULT_DURATION) #

[View source]

Instance Method Detail

def lost? : Bool #

[View source]

def start : Nil #

[View source]

def stop : Nil #

[View source]