SERVER-14052 Make the autosplitter work with less then three splitpoints

SplitVector is always removing the first splitpoint from the splitvector
before returning it, thus the autsplitter should try to split even when
the splitvector contains just one splitpoint.

2 files changed, 54 insertions, 3 deletions

diff --git a/jstests/sharding/autosplit_low_cardinality.js b/jstests/sharding/autosplit_low_cardinality.js
new file mode 100644
index 00000000000..c3b91a6a7c7
--- /dev/null
+++ b/jstests/sharding/autosplit_low_cardinality.js
@@ -0,0 +1,53 @@
+/**
+ * Test the autosplitter when a collection has very low cardinality
+ *
+ * @tags: [requires_fcv_44]
+ */
+
+(function() {
+'use strict';
+load('jstests/sharding/autosplit_include.js');
+
+var st = new ShardingTest({
+    name: "low_cardinality",
+    other: {enableAutoSplit: true, chunkSize: 1},
+});
+
+assert.commandWorked(st.s.adminCommand({enablesharding: "test"}));
+assert.commandWorked(st.s.adminCommand({shardcollection: "test.foo", key: {sk: 1}}));
+
+const bigString = "X".repeat(1024 * 1024 / 4);  // 250 KB
+
+var coll = st.getDB("test").getCollection("foo");
+
+// Insert $numDocs documents into the collection under $key.
+// Each document contains a string of 250KB
+// waits for any ongoing splits to finish, and then prints some information
+// about the collection's chunks
+function insertBigDocsWithKey(key, numDocs) {
+    var bulk = coll.initializeUnorderedBulkOp();
+    for (var i = 0; i < numDocs; i++) {
+        bulk.insert({sk: key, sub: i, bs: bigString});
+    }
+    assert.commandWorked(bulk.execute());
+}
+
+function numChunks() {
+    return st.config.chunks.count({"ns": "test.foo"});
+}
+
+// Accumulate ~1MB of documents under -10 and +10
+insertBigDocsWithKey(-10, 4);
+insertBigDocsWithKey(10, 4);
+waitForOngoingChunkSplits(st);
+
+// At least one split should have been performed
+assert.gte(numChunks(), 2, "Number of chunks is less then 2, no split have been perfomed");
+
+insertBigDocsWithKey(20, 4);
+waitForOngoingChunkSplits(st);
+// An additional split should have been performed
+assert.gte(numChunks(), 3, "Number of chunks must be at least 3");
+
+st.stop();
+})();
diff --git a/src/mongo/db/s/chunk_splitter.cpp b/src/mongo/db/s/chunk_splitter.cpp
index c7dd1e22250..33235a3be68 100644
--- a/src/mongo/db/s/chunk_splitter.cpp
+++ b/src/mongo/db/s/chunk_splitter.cpp
@@ -328,15 +328,13 @@ void ChunkSplitter::_runAutosplit(std::shared_ptr<ChunkSplitStateDriver> chunkSp
                                                        boost::none,
                                                        maxChunkSizeBytes));
 
-        if (splitPoints.size() <= 1) {
+        if (splitPoints.empty()) {
             LOG(1)
                 << "ChunkSplitter attempted split but not enough split points were found for chunk "
                 << redact(chunk.toString());
             // Reset our size estimate that we had prior to splitVector to 0, while still counting
             // the bytes that have been written in parallel to this split task
             chunkSplitStateDriver->abandonPrepare();
-            // No split points means there isn't enough data to split on; 1 split point means we
-            // have between half the chunk size to full chunk size so there is no need to split yet
             return;
         }