public void printLocalityAndDispersionForCurrentPlan(
Map<String, Map<String, Float>> regionLocalityMap) throws IOException {
SnapshotOfRegionAssignmentFromMeta snapshot = this.getRegionAssignmentSnapshot();
FavoredNodesPlan assignmentPlan = snapshot.getExistingAssignmentPlan();
Set<TableName> tables = snapshot.getTableSet();
Map<TableName, List<HRegionInfo>> tableToRegionsMap = snapshot.getTableToRegionMap();
for (TableName table : tables) {
float[] locality = new float[3];
if (!this.targetTableSet.isEmpty() && !this.targetTableSet.contains(table)) {
continue;
}
List<HRegionInfo> regions = tableToRegionsMap.get(table);
for (HRegionInfo currentRegion : regions) {
Map<String, Float> regionLocality = regionLocalityMap.get(currentRegion.getEncodedName());
if (regionLocality == null) {
continue;
}
List<ServerName> servers = assignmentPlan.getFavoredNodes(currentRegion);
if (servers != null) {
int i = 0;
for (FavoredNodesPlan.Position p : FavoredNodesPlan.Position.values()) {
ServerName server = servers.get(p.ordinal());
Float currentLocality = 0f;
if (servers != null) {
currentLocality = regionLocality.get(server.getHostname());
if (currentLocality == null) {
currentLocality = 0f;
}
locality[i] += currentLocality;
}
i++;
}
}
}
for (int i = 0; i < locality.length; i++) {
String copy = null;
if (i == 0) {
copy = "primary";
} else if (i == 1) {
copy = "secondary";
} else if (i == 2) {
copy = "tertiary";
}
float avgLocality = 100 * locality[i] / regions.size();
LOG.info(
"For Table: "
+ table
+ " ; #Total Regions: "
+ regions.size()
+ " ; The average locality for "
+ copy
+ " is "
+ avgLocality
+ " %");
}
printDispersionScores(table, snapshot, regions.size(), null, false);
}
}
/**
* Compares two plans and check whether the locality dropped or increased (prints the information
* as a string) also prints the baseline locality
*
* @param movesPerTable - how many primary regions will move per table
* @param regionLocalityMap - locality map from FS
* @param newPlan - new assignment plan
* @throws IOException
*/
public void checkDifferencesWithOldPlan(
Map<TableName, Integer> movesPerTable,
Map<String, Map<String, Float>> regionLocalityMap,
FavoredNodesPlan newPlan)
throws IOException {
// localities for primary, secondary and tertiary
SnapshotOfRegionAssignmentFromMeta snapshot = this.getRegionAssignmentSnapshot();
FavoredNodesPlan oldPlan = snapshot.getExistingAssignmentPlan();
Set<TableName> tables = snapshot.getTableSet();
Map<TableName, List<HRegionInfo>> tableToRegionsMap = snapshot.getTableToRegionMap();
for (TableName table : tables) {
float[] deltaLocality = new float[3];
float[] locality = new float[3];
if (!this.targetTableSet.isEmpty() && !this.targetTableSet.contains(table)) {
continue;
}
List<HRegionInfo> regions = tableToRegionsMap.get(table);
System.out.println("==================================================");
System.out.println("Assignment Plan Projection Report For Table: " + table);
System.out.println("\t Total regions: " + regions.size());
System.out.println(
"\t"
+ movesPerTable.get(table)
+ " primaries will move due to their primary has changed");
for (HRegionInfo currentRegion : regions) {
Map<String, Float> regionLocality = regionLocalityMap.get(currentRegion.getEncodedName());
if (regionLocality == null) {
continue;
}
List<ServerName> oldServers = oldPlan.getFavoredNodes(currentRegion);
List<ServerName> newServers = newPlan.getFavoredNodes(currentRegion);
if (newServers != null && oldServers != null) {
int i = 0;
for (FavoredNodesPlan.Position p : FavoredNodesPlan.Position.values()) {
ServerName newServer = newServers.get(p.ordinal());
ServerName oldServer = oldServers.get(p.ordinal());
Float oldLocality = 0f;
if (oldServers != null) {
oldLocality = regionLocality.get(oldServer.getHostname());
if (oldLocality == null) {
oldLocality = 0f;
}
locality[i] += oldLocality;
}
Float newLocality = regionLocality.get(newServer.getHostname());
if (newLocality == null) {
newLocality = 0f;
}
deltaLocality[i] += newLocality - oldLocality;
i++;
}
}
}
DecimalFormat df = new java.text.DecimalFormat("#.##");
for (int i = 0; i < deltaLocality.length; i++) {
System.out.print("\t\t Baseline locality for ");
if (i == 0) {
System.out.print("primary ");
} else if (i == 1) {
System.out.print("secondary ");
} else if (i == 2) {
System.out.print("tertiary ");
}
System.out.println(df.format(100 * locality[i] / regions.size()) + "%");
System.out.print("\t\t Locality will change with the new plan: ");
System.out.println(df.format(100 * deltaLocality[i] / regions.size()) + "%");
}
System.out.println("\t Baseline dispersion");
printDispersionScores(table, snapshot, regions.size(), null, true);
System.out.println("\t Projected dispersion");
printDispersionScores(table, snapshot, regions.size(), newPlan, true);
}
}
/**
* Return the number of regions based on the position (primary/secondary/ tertiary) assigned to
* their favored nodes
*
* @param position
* @return the number of regions
*/
int getNumRegionsOnFavoredNodeByPosition(FavoredNodesPlan.Position position) {
return favoredNodes[position.ordinal()];
}
public void fillUp(
TableName tableName,
SnapshotOfRegionAssignmentFromMeta snapshot,
Map<String, Map<String, Float>> regionLocalityMap) {
// Set the table name
this.tableName = tableName;
// Get all the regions for this table
List<HRegionInfo> regionInfoList = snapshot.getTableToRegionMap().get(tableName);
// Get the total region num for the current table
this.totalRegions = regionInfoList.size();
// Get the existing assignment plan
FavoredNodesPlan favoredNodesAssignment = snapshot.getExistingAssignmentPlan();
// Get the region to region server mapping
Map<HRegionInfo, ServerName> currentAssignment = snapshot.getRegionToRegionServerMap();
// Initialize the server to its hosing region counter map
Map<ServerName, Integer> serverToHostingRegionCounterMap = new HashMap<ServerName, Integer>();
Map<ServerName, Integer> primaryRSToRegionCounterMap = new HashMap<ServerName, Integer>();
Map<ServerName, Set<ServerName>> primaryToSecTerRSMap =
new HashMap<ServerName, Set<ServerName>>();
// Check the favored nodes and its locality information
// Also keep tracker of the most loaded and least loaded region servers
for (HRegionInfo region : regionInfoList) {
try {
ServerName currentRS = currentAssignment.get(region);
// Handle unassigned regions
if (currentRS == null) {
unAssignedRegionsList.add(region);
continue;
}
// Keep updating the server to is hosting region counter map
Integer hostRegionCounter = serverToHostingRegionCounterMap.get(currentRS);
if (hostRegionCounter == null) {
hostRegionCounter = Integer.valueOf(0);
}
hostRegionCounter = hostRegionCounter.intValue() + 1;
serverToHostingRegionCounterMap.put(currentRS, hostRegionCounter);
// Get the favored nodes from the assignment plan and verify it.
List<ServerName> favoredNodes = favoredNodesAssignment.getFavoredNodes(region);
if (favoredNodes == null
|| favoredNodes.size() != FavoredNodeAssignmentHelper.FAVORED_NODES_NUM) {
regionsWithoutValidFavoredNodes.add(region);
continue;
}
// Get the primary, secondary and tertiary region server
ServerName primaryRS = favoredNodes.get(FavoredNodesPlan.Position.PRIMARY.ordinal());
ServerName secondaryRS = favoredNodes.get(FavoredNodesPlan.Position.SECONDARY.ordinal());
ServerName tertiaryRS = favoredNodes.get(FavoredNodesPlan.Position.TERTIARY.ordinal());
// Update the primary rs to its region set map
Integer regionCounter = primaryRSToRegionCounterMap.get(primaryRS);
if (regionCounter == null) {
regionCounter = Integer.valueOf(0);
}
regionCounter = regionCounter.intValue() + 1;
primaryRSToRegionCounterMap.put(primaryRS, regionCounter);
// Update the primary rs to secondary and tertiary rs map
Set<ServerName> secAndTerSet = primaryToSecTerRSMap.get(primaryRS);
if (secAndTerSet == null) {
secAndTerSet = new HashSet<ServerName>();
}
secAndTerSet.add(secondaryRS);
secAndTerSet.add(tertiaryRS);
primaryToSecTerRSMap.put(primaryRS, secAndTerSet);
// Get the position of the current region server in the favored nodes list
FavoredNodesPlan.Position favoredNodePosition =
FavoredNodesPlan.getFavoredServerPosition(favoredNodes, currentRS);
// Handle the non favored assignment.
if (favoredNodePosition == null) {
nonFavoredAssignedRegionList.add(region);
continue;
}
// Increase the favored nodes assignment.
this.favoredNodes[favoredNodePosition.ordinal()]++;
totalFavoredAssignments++;
// Summary the locality information for each favored nodes
if (regionLocalityMap != null) {
// Set the enforce locality as true;
this.enforceLocality = true;
// Get the region degree locality map
Map<String, Float> regionDegreeLocalityMap =
regionLocalityMap.get(region.getEncodedName());
if (regionDegreeLocalityMap == null) {
continue; // ignore the region which doesn't have any store files.
}
// Get the locality summary for each favored nodes
for (FavoredNodesPlan.Position p : FavoredNodesPlan.Position.values()) {
ServerName favoredNode = favoredNodes.get(p.ordinal());
// Get the locality for the current favored nodes
Float locality = regionDegreeLocalityMap.get(favoredNode.getHostname());
if (locality != null) {
this.favoredNodesLocalitySummary[p.ordinal()] += locality;
}
}
// Get the locality summary for the current region server
Float actualLocality = regionDegreeLocalityMap.get(currentRS.getHostname());
if (actualLocality != null) {
this.actualLocalitySummary += actualLocality;
}
}
} catch (Exception e) {
LOG.error(
"Cannot verify the region assignment for region "
+ ((region == null) ? " null " : region.getRegionNameAsString())
+ "because of "
+ e);
}
}
float dispersionScoreSummary = 0;
float dispersionNumSummary = 0;
// Calculate the secondary score for each primary region server
for (Map.Entry<ServerName, Integer> entry : primaryRSToRegionCounterMap.entrySet()) {
ServerName primaryRS = entry.getKey();
Integer regionsOnPrimary = entry.getValue();
// Process the dispersion number and score
float dispersionScore = 0;
int dispersionNum = 0;
if (primaryToSecTerRSMap.get(primaryRS) != null && regionsOnPrimary.intValue() != 0) {
dispersionNum = primaryToSecTerRSMap.get(primaryRS).size();
dispersionScore = dispersionNum / ((float) regionsOnPrimary.intValue() * 2);
}
// Update the max dispersion score
if (dispersionScore > this.maxDispersionScore) {
this.maxDispersionScoreServerSet.clear();
this.maxDispersionScoreServerSet.add(primaryRS);
this.maxDispersionScore = dispersionScore;
} else if (dispersionScore == this.maxDispersionScore) {
this.maxDispersionScoreServerSet.add(primaryRS);
}
// Update the max dispersion num
if (dispersionNum > this.maxDispersionNum) {
this.maxDispersionNumServerSet.clear();
this.maxDispersionNumServerSet.add(primaryRS);
this.maxDispersionNum = dispersionNum;
} else if (dispersionNum == this.maxDispersionNum) {
this.maxDispersionNumServerSet.add(primaryRS);
}
// Update the min dispersion score
if (dispersionScore < this.minDispersionScore) {
this.minDispersionScoreServerSet.clear();
this.minDispersionScoreServerSet.add(primaryRS);
this.minDispersionScore = dispersionScore;
} else if (dispersionScore == this.minDispersionScore) {
this.minDispersionScoreServerSet.add(primaryRS);
}
// Update the min dispersion num
if (dispersionNum < this.minDispersionNum) {
this.minDispersionNumServerSet.clear();
this.minDispersionNumServerSet.add(primaryRS);
this.minDispersionNum = dispersionNum;
} else if (dispersionNum == this.minDispersionNum) {
this.minDispersionNumServerSet.add(primaryRS);
}
dispersionScoreSummary += dispersionScore;
dispersionNumSummary += dispersionNum;
}
// Update the avg dispersion score
if (primaryRSToRegionCounterMap.keySet().size() != 0) {
this.avgDispersionScore =
dispersionScoreSummary / (float) primaryRSToRegionCounterMap.keySet().size();
this.avgDispersionNum =
dispersionNumSummary / (float) primaryRSToRegionCounterMap.keySet().size();
}
// Fill up the most loaded and least loaded region server information
for (Map.Entry<ServerName, Integer> entry : serverToHostingRegionCounterMap.entrySet()) {
ServerName currentRS = entry.getKey();
int hostRegionCounter = entry.getValue().intValue();
// Update the most loaded region server list and maxRegionsOnRS
if (hostRegionCounter > this.maxRegionsOnRS) {
maxRegionsOnRS = hostRegionCounter;
this.mostLoadedRSSet.clear();
this.mostLoadedRSSet.add(currentRS);
} else if (hostRegionCounter == this.maxRegionsOnRS) {
this.mostLoadedRSSet.add(currentRS);
}
// Update the least loaded region server list and minRegionsOnRS
if (hostRegionCounter < this.minRegionsOnRS) {
this.minRegionsOnRS = hostRegionCounter;
this.leastLoadedRSSet.clear();
this.leastLoadedRSSet.add(currentRS);
} else if (hostRegionCounter == this.minRegionsOnRS) {
this.leastLoadedRSSet.add(currentRS);
}
}
// and total region servers
this.totalRegionServers = serverToHostingRegionCounterMap.keySet().size();
this.avgRegionsOnRS =
(totalRegionServers == 0) ? 0 : (totalRegions / (float) totalRegionServers);
// Set the isFilledUp as true
isFilledUp = true;
}
public void print(boolean isDetailMode) {
if (!isFilledUp) {
System.err.println("[Error] Region assignment verfication report" + "hasn't been filled up");
}
DecimalFormat df = new java.text.DecimalFormat("#.##");
// Print some basic information
System.out.println(
"Region Assignment Verification for Table: "
+ tableName
+ "\n\tTotal regions : "
+ totalRegions);
// Print the number of regions on each kinds of the favored nodes
System.out.println("\tTotal regions on favored nodes " + totalFavoredAssignments);
for (FavoredNodesPlan.Position p : FavoredNodesPlan.Position.values()) {
System.out.println(
"\t\tTotal regions on " + p.toString() + " region servers: " + favoredNodes[p.ordinal()]);
}
// Print the number of regions in each kinds of invalid assignment
System.out.println("\tTotal unassigned regions: " + unAssignedRegionsList.size());
if (isDetailMode) {
for (HRegionInfo region : unAssignedRegionsList) {
System.out.println("\t\t" + region.getRegionNameAsString());
}
}
System.out.println(
"\tTotal regions NOT on favored nodes: " + nonFavoredAssignedRegionList.size());
if (isDetailMode) {
for (HRegionInfo region : nonFavoredAssignedRegionList) {
System.out.println("\t\t" + region.getRegionNameAsString());
}
}
System.out.println(
"\tTotal regions without favored nodes: " + regionsWithoutValidFavoredNodes.size());
if (isDetailMode) {
for (HRegionInfo region : regionsWithoutValidFavoredNodes) {
System.out.println("\t\t" + region.getRegionNameAsString());
}
}
// Print the locality information if enabled
if (this.enforceLocality && totalRegions != 0) {
// Print the actual locality for this table
float actualLocality = 100 * this.actualLocalitySummary / (float) totalRegions;
System.out.println("\n\tThe actual avg locality is " + df.format(actualLocality) + " %");
// Print the expected locality if regions are placed on the each kinds of
// favored nodes
for (FavoredNodesPlan.Position p : FavoredNodesPlan.Position.values()) {
float avgLocality = 100 * (favoredNodesLocalitySummary[p.ordinal()] / (float) totalRegions);
System.out.println(
"\t\tThe expected avg locality if all regions"
+ " on the "
+ p.toString()
+ " region servers: "
+ df.format(avgLocality)
+ " %");
}
}
// Print the region balancing information
System.out.println("\n\tTotal hosting region servers: " + totalRegionServers);
// Print the region balance information
if (totalRegionServers != 0) {
System.out.println(
"\tAvg dispersion num: "
+ df.format(avgDispersionNum)
+ " hosts;\tMax dispersion num: "
+ df.format(maxDispersionNum)
+ " hosts;\tMin dispersion num: "
+ df.format(minDispersionNum)
+ " hosts;");
System.out.println(
"\t\tThe number of the region servers with the max"
+ " dispersion num: "
+ this.maxDispersionNumServerSet.size());
if (isDetailMode) {
printHServerAddressSet(maxDispersionNumServerSet);
}
System.out.println(
"\t\tThe number of the region servers with the min"
+ " dispersion num: "
+ this.minDispersionNumServerSet.size());
if (isDetailMode) {
printHServerAddressSet(maxDispersionNumServerSet);
}
System.out.println(
"\tAvg dispersion score: "
+ df.format(avgDispersionScore)
+ ";\tMax dispersion score: "
+ df.format(maxDispersionScore)
+ ";\tMin dispersion score: "
+ df.format(minDispersionScore)
+ ";");
System.out.println(
"\t\tThe number of the region servers with the max"
+ " dispersion score: "
+ this.maxDispersionScoreServerSet.size());
if (isDetailMode) {
printHServerAddressSet(maxDispersionScoreServerSet);
}
System.out.println(
"\t\tThe number of the region servers with the min"
+ " dispersion score: "
+ this.minDispersionScoreServerSet.size());
if (isDetailMode) {
printHServerAddressSet(minDispersionScoreServerSet);
}
System.out.println(
"\tAvg regions/region server: "
+ df.format(avgRegionsOnRS)
+ ";\tMax regions/region server: "
+ maxRegionsOnRS
+ ";\tMin regions/region server: "
+ minRegionsOnRS
+ ";");
// Print the details about the most loaded region servers
System.out.println(
"\t\tThe number of the most loaded region servers: " + mostLoadedRSSet.size());
if (isDetailMode) {
printHServerAddressSet(mostLoadedRSSet);
}
// Print the details about the least loaded region servers
System.out.println(
"\t\tThe number of the least loaded region servers: " + leastLoadedRSSet.size());
if (isDetailMode) {
printHServerAddressSet(leastLoadedRSSet);
}
}
System.out.println("==============================");
}
