Examples with NormalizedResourceOffer org.apache.storm.scheduler.resource.normalization.NormalizedResourceOffer used on opensource projects

Example 11 with NormalizedResourceOffer

use of org.apache.storm.scheduler.resource.normalization.NormalizedResourceOffer in project storm by apache.

the class RasBlacklistStrategy method releaseBlacklistWhenNeeded.

@Override
protected Set<String> releaseBlacklistWhenNeeded(Cluster cluster, final List<String> blacklistedNodeIds) {
    LOG.info("RAS We have {} nodes blacklisted...", blacklistedNodeIds.size());
    Set<String> readyToRemove = new HashSet<>();
    if (blacklistedNodeIds.size() > 0) {
        int availableSlots = cluster.getNonBlacklistedAvailableSlots(blacklistedNodeIds).size();
        int neededSlots = 0;
        NormalizedResourceOffer available = cluster.getNonBlacklistedClusterAvailableResources(blacklistedNodeIds);
        NormalizedResourceOffer needed = new NormalizedResourceOffer();
        for (TopologyDetails td : cluster.getTopologies()) {
            if (cluster.needsSchedulingRas(td)) {
                int slots = 0;
                try {
                    slots = ServerUtils.getEstimatedWorkerCountForRasTopo(td.getConf(), td.getTopology());
                } catch (InvalidTopologyException e) {
                    LOG.warn("Could not guess the number of slots needed for {}", td.getName(), e);
                }
                int assignedSlots = cluster.getAssignedNumWorkers(td);
                int tdSlotsNeeded = slots - assignedSlots;
                neededSlots += tdSlotsNeeded;
                NormalizedResourceRequest resources = td.getApproximateTotalResources();
                needed.add(resources);
                LOG.warn("{} needs to be scheduled with {} and {} slots", td.getName(), resources, tdSlotsNeeded);
            }
        }
        // Now we need to free up some resources...
        Map<String, SupervisorDetails> availableSupervisors = cluster.getSupervisors();
        NormalizedResourceOffer shortage = new NormalizedResourceOffer(needed);
        shortage.remove(available, cluster.getResourceMetrics());
        int shortageSlots = neededSlots - availableSlots;
        LOG.debug("Need {} and {} slots.", needed, neededSlots);
        LOG.debug("Available {} and {} slots.", available, availableSlots);
        LOG.debug("Shortage {} and {} slots.", shortage, shortageSlots);
        if (shortage.areAnyOverZero() || shortageSlots > 0) {
            LOG.info("Need {} and {} slots more. Releasing some blacklisted nodes to cover it.", shortage, shortageSlots);
            // release earliest blacklist - but release all supervisors on a given blacklisted host.
            Map<String, Set<String>> hostToSupervisorIds = createHostToSupervisorMap(blacklistedNodeIds, cluster);
            for (Set<String> supervisorIds : hostToSupervisorIds.values()) {
                for (String supervisorId : supervisorIds) {
                    SupervisorDetails sd = availableSupervisors.get(supervisorId);
                    if (sd != null) {
                        NormalizedResourcesWithMemory sdAvailable = cluster.getAvailableResources(sd);
                        int sdAvailableSlots = cluster.getAvailablePorts(sd).size();
                        readyToRemove.add(supervisorId);
                        shortage.remove(sdAvailable, cluster.getResourceMetrics());
                        shortageSlots -= sdAvailableSlots;
                        LOG.info("Releasing {} with {} and {} slots leaving {} and {} slots to go", supervisorId, sdAvailable, sdAvailableSlots, shortage, shortageSlots);
                    }
                }
                // make sure we've handled all supervisors on the host before we break
                if (!shortage.areAnyOverZero() && shortageSlots <= 0) {
                    // we have enough resources now...
                    break;
                }
            }
        }
    }
    return readyToRemove;
}