somandal commented on code in PR #15618: URL: https://github.com/apache/pinot/pull/15618#discussion_r2076151354
########## pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/rebalance/TableRebalancer.java: ########## @@ -1352,26 +1380,55 @@ static boolean isExternalViewConverged(String tableNameWithType, Map<String, Map<String, String>> externalViewSegmentStates, Map<String, Map<String, String>> idealStateSegmentStates, boolean lowDiskMode, boolean bestEfforts, @Nullable Set<String> segmentsToMonitor) { - return isExternalViewConverged(tableNameWithType, externalViewSegmentStates, idealStateSegmentStates, lowDiskMode, - bestEfforts, segmentsToMonitor, LOGGER); + return getNumRemainingSegmentsToProcess(tableNameWithType, externalViewSegmentStates, idealStateSegmentStates, + lowDiskMode, bestEfforts, segmentsToMonitor, LOGGER, true) == 0; } /** - * NOTE: - * Only check the segments in the IdealState and being monitored. Extra segments in ExternalView are ignored because - * they are not managed by the rebalancer. - * For each segment checked: - * - In regular mode, it is okay to have extra instances in ExternalView as long as the instance states in IdealState - * are reached. - * - In low disk mode, instance states in ExternalView must match IdealState to ensure the segments are deleted from - * server before moving to the next assignment. - * For ERROR state in ExternalView, if using best-efforts, log a warning and treat it as good state; if not, throw an - * exception to abort the rebalance because we are not able to get out of the ERROR state. + * Check if the external view has converged to the ideal state. See `getNumRemainingSegmentsToProcess` for details on + * how the convergence is determined. */ private static boolean isExternalViewConverged(String tableNameWithType, Map<String, Map<String, String>> externalViewSegmentStates, Map<String, Map<String, String>> idealStateSegmentStates, boolean lowDiskMode, boolean bestEfforts, @Nullable Set<String> segmentsToMonitor, Logger tableRebalanceLogger) { + return getNumRemainingSegmentsToProcess(tableNameWithType, externalViewSegmentStates, idealStateSegmentStates, + lowDiskMode, bestEfforts, segmentsToMonitor, tableRebalanceLogger, true) == 0; + } + + @VisibleForTesting + static int getNumRemainingSegmentsToProcess(String tableNameWithType, + Map<String, Map<String, String>> externalViewSegmentStates, + Map<String, Map<String, String>> idealStateSegmentStates, boolean lowDiskMode, boolean bestEfforts, + @Nullable Set<String> segmentsToMonitor) { + return getNumRemainingSegmentsToProcess(tableNameWithType, externalViewSegmentStates, idealStateSegmentStates, + lowDiskMode, bestEfforts, segmentsToMonitor, LOGGER, false); + } + + /** + * Count the number of segments that are not in the expected state. If `earlyReturn=true` it returns 1 as soon as + * the count becomes non-zero. This is used to check whether the ExternalView has converged to the IdealState. The + * method checks the following: + * Only the segments in the IdealState and being monitored. Extra segments in ExternalView are ignored + * because they are not managed by the rebalancer. + * For each segment, go through instances in the instance map from IdealState and compare it with the one in + * ExternalView, and increment the number of remaining segments to process if: + * - The instance appears in IS instance map, but there is no instance map in EV, unless the IS instance state is + * OFFLINE + * - The instance appears in IS instance map is not in the EV instance map, unless the IS instance state is OFFLINE + * - The instance has different states between IS and EV instance map, unless the IS instance state is OFFLINE Review Comment: I think we almost never set a segment to OFFLINE state in IS, except when we reset the segment. If the segment is OFFLINE in IS we assume that segment is not needed to be online as IS is the expected state the segment should be in. To remove segments, we delete them or remove the server instance for that segment. OFFLINE in EV but ONLINE in IS is different - indicating that the server might be down or the segment hasn't completed state transition change yet. does this help? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org For additional commands, e-mail: commits-h...@pinot.apache.org