keith-turner commented on code in PR #3369:
URL: https://github.com/apache/accumulo/pull/3369#discussion_r1185226923
##########
hadoop-mapreduce/src/main/java/org/apache/accumulo/hadoopImpl/mapred/AccumuloRecordReader.java:
##########
@@ -348,14 +355,45 @@ public static InputSplit[] getSplits(JobConf job,
Class<?> callingClass) throws
// tablets... so clear it
tl.invalidateCache();
- while (!tl.binRanges(context, ranges, binnedRanges).isEmpty()) {
- context.requireNotDeleted(tableId);
- context.requireNotOffline(tableId, tableName);
- binnedRanges.clear();
- log.warn("Unable to locate bins for specified ranges.
Retrying.");
- // sleep randomly between 100 and 200 ms
- sleepUninterruptibly(100 + random.nextInt(100),
TimeUnit.MILLISECONDS);
- tl.invalidateCache();
+ if (InputConfigurator.getConsistencyLevel(callingClass, job)
+ == ConsistencyLevel.IMMEDIATE) {
+ while (!tl.binRanges(context, ranges, binnedRanges).isEmpty()) {
+ context.requireNotDeleted(tableId);
+ context.requireNotOffline(tableId, tableName);
+ binnedRanges.clear();
+ log.warn("Unable to locate bins for specified ranges.
Retrying.");
+ // sleep randomly between 100 and 200 ms
+ sleepUninterruptibly(100 + random.nextInt(100),
TimeUnit.MILLISECONDS);
+ tl.invalidateCache();
+ }
+ } else {
+ Map<String,Map<KeyExtent,List<Range>>> unhostedRanges = new
HashMap<>();
+ unhostedRanges.put("", new HashMap<>());
+ BiConsumer<CachedTablet,Range> consumer = (ct, r) -> {
+ unhostedRanges.get("").computeIfAbsent(ct.getExtent(), k ->
new ArrayList<>())
+ .add(r);
+ };
+ List<Range> failures =
+ tl.findTablets(context, ranges, consumer,
LocationNeed.NOT_REQUIRED);
+
+ Retry retry = Retry.builder().infiniteRetries().retryAfter(100,
MILLISECONDS)
+ .incrementBy(100, MILLISECONDS).maxWait(2,
SECONDS).backOffFactor(1.5)
+ .logInterval(3, MINUTES).createRetry();
+
+ while (!failures.isEmpty()) {
+
Review Comment:
Need to check if the table still exists when failures is non empty. Things
end up in failures because it could not find a tablet that overlapped the
range. This could be caused by split, merge, or table deletion.
##########
hadoop-mapreduce/src/main/java/org/apache/accumulo/hadoopImpl/mapreduce/AccumuloRecordReader.java:
##########
@@ -381,14 +388,46 @@ public static List<InputSplit> getSplits(JobContext
context, Class<?> callingCla
// tables tablets... so clear it
tl.invalidateCache();
- while (!tl.binRanges(clientContext, ranges,
binnedRanges).isEmpty()) {
- clientContext.requireNotDeleted(tableId);
- clientContext.requireNotOffline(tableId, tableName);
- binnedRanges.clear();
- log.warn("Unable to locate bins for specified ranges.
Retrying.");
- // sleep randomly between 100 and 200 ms
- sleepUninterruptibly(100 + random.nextInt(100),
TimeUnit.MILLISECONDS);
- tl.invalidateCache();
+ if (InputConfigurator.getConsistencyLevel(callingClass,
context.getConfiguration())
+ == ConsistencyLevel.IMMEDIATE) {
+ while (!tl.binRanges(clientContext, ranges,
binnedRanges).isEmpty()) {
+ clientContext.requireNotDeleted(tableId);
+ clientContext.requireNotOffline(tableId, tableName);
+ binnedRanges.clear();
+ log.warn("Unable to locate bins for specified ranges.
Retrying.");
+ // sleep randomly between 100 and 200 ms
+ sleepUninterruptibly(100 + random.nextInt(100),
TimeUnit.MILLISECONDS);
+ tl.invalidateCache();
+ }
+ } else {
+ Map<String,Map<KeyExtent,List<Range>>> unhostedRanges = new
HashMap<>();
+ unhostedRanges.put("", new HashMap<>());
+ BiConsumer<CachedTablet,Range> consumer = (ct, r) -> {
+ unhostedRanges.get("").computeIfAbsent(ct.getExtent(), k ->
new ArrayList<>())
+ .add(r);
+ };
+ List<Range> failures =
+ tl.findTablets(clientContext, ranges, consumer,
LocationNeed.NOT_REQUIRED);
+
+ Retry retry = Retry.builder().infiniteRetries().retryAfter(100,
MILLISECONDS)
+ .incrementBy(100, MILLISECONDS).maxWait(2,
SECONDS).backOffFactor(1.5)
+ .logInterval(3, MINUTES).createRetry();
+
+ while (!failures.isEmpty()) {
+
Review Comment:
also need table deletion check
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
