[KYLIN-2973] Potential issue of not atomically update cube instance map

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: v2.6.0
• Assignee: Zhong Yanghong
• Reporter: Zhong Yanghong
• Created: 27/Oct/17 09:16
• Updated: 24/Dec/18 07:48
• Resolved: 24/Dec/18 07:48

Description

P1

        try {
            getStore().putResource(cube.getResourcePath(), cube, CUBE_SERIALIZER);
        } catch (IllegalStateException ise) {
            logger.warn("Write conflict to update cube " + cube.getName() + " at try " + retry + ", will retry...");
            if (retry >= 7) {
                logger.error("Retried 7 times till got error, abandoning...", ise);
                throw ise;
            }

            cube = reloadCubeLocal(cube.getName());
            update.setCubeInstance(cube);
            retry++;
            cube = updateCubeWithRetry(update, retry);
        }

P2

        if (toRemoveResources.size() > 0) {
            for (String resource : toRemoveResources) {
                try {
                    getStore().deleteResource(resource);
                } catch (IOException ioe) {
                    logger.error("Failed to delete resource " + toRemoveResources.toString());
                }
            }
        }

P3

        cubeMap.put(cube.getName(), cube);

There's a chance like:

1. Thread t1, goes into P2;
2. Then Thread t2, goes into P1, P2, P3; the cube instance in the map will be updated by t2
3. Then Thread t1 goes into P3; the cube instance in the map will be updated by t1, which is not correct

Issue Links

• relates to: KYLIN-3085 CubeManager.updateCube() must not update the cached CubeInstance