Examples with HighAvailabilityStoreFailureException org.neo4j.kernel.ha.store.HighAvailabilityStoreFailureException used on opensource projects

Example 1 with HighAvailabilityStoreFailureException

use of org.neo4j.kernel.ha.store.HighAvailabilityStoreFailureException in project neo4j by neo4j.

the class HighAvailabilityModeSwitcher method switchToSlave.

private void switchToSlave() {
    /*
         * This is purely defensive and should never trigger. There was a race where the switch to slave task would
         * start after this instance was elected master and the task would constantly try to change as slave
         * for itself, never cancelling. This now should not be possible, since we cancel the task and wait for it
         * to complete, all in a single thread executor. However, this is a check worth doing because if this
         * condition slips through via some other code path it can cause trouble.
         */
    if (getServerId(availableMasterId).equals(instanceId)) {
        msgLog.error("I (" + me + ") tried to switch to slave for myself as master (" + availableMasterId + ")");
        return;
    }
    final AtomicLong wait = new AtomicLong();
    final CancellationHandle cancellationHandle = new CancellationHandle();
    startModeSwitching(new Runnable() {

        @Override
        public void run() {
            if (currentTargetState != HighAvailabilityMemberState.TO_SLAVE) {
                // Already switched - this can happen if a second master becomes available while waiting
                return;
            }
            if (cancellationHandle.cancellationRequested()) {
                msgLog.info("Switch to slave cancelled on start.");
                return;
            }
            componentSwitcher.switchToSlave();
            try {
                if (cancellationHandle.cancellationRequested()) {
                    msgLog.info("Switch to slave cancelled before ha communication started.");
                    return;
                }
                haCommunicationLife.shutdown();
                haCommunicationLife = new LifeSupport();
                // it is important for availableMasterId to be re-read on every attempt so that
                // slave switching would not result in an infinite loop with wrong/stale availableMasterId
                URI resultingSlaveHaURI = switchToSlave.switchToSlave(haCommunicationLife, me, availableMasterId, cancellationHandle);
                if (resultingSlaveHaURI == null) {
                    /*
                         * null slave uri means the task was cancelled. The task then must simply terminate and
                         * have no side effects.
                         */
                    msgLog.info("Switch to slave is effectively cancelled");
                } else {
                    slaveHaURI = resultingSlaveHaURI;
                    canAskForElections.set(true);
                }
            } catch (HighAvailabilityStoreFailureException e) {
                userLog.error("UNABLE TO START UP AS SLAVE: %s", e.getMessage());
                msgLog.error("Unable to start up as slave", e);
                clusterMemberAvailability.memberIsUnavailable(SLAVE);
                ClusterClient clusterClient = HighAvailabilityModeSwitcher.this.clusterClient;
                try {
                    // TODO I doubt this actually works
                    clusterClient.leave();
                    clusterClient.stop();
                    haCommunicationLife.shutdown();
                } catch (Throwable t) {
                    msgLog.error("Unable to stop cluster client", t);
                }
                modeSwitcherExecutor.schedule(this, 5, TimeUnit.SECONDS);
            } catch (MismatchingStoreIdException e) {
                // Try again immediately, the place that threw it have already treated the db
                // as branched and so a new attempt will have this slave copy a new store from master.
                run();
            } catch (Throwable t) {
                msgLog.error("Error while trying to switch to slave", t);
                // Try again later
                // Exponential backoff
                wait.set(1 + wait.get() * 2);
                // Wait maximum 5 minutes
                wait.set(Math.min(wait.get(), 5 * 60));
                modeSwitcherFuture = modeSwitcherExecutor.schedule(this, wait.get(), TimeUnit.SECONDS);
                msgLog.info("Attempting to switch to slave in %ds", wait.get());
            }
        }
    }, cancellationHandle);
}