Write metadata cache data to mappings _meta with refresh time update (#…

…805) (#840) * [0.5-nexus] Write mock metadata cache data to mappings _meta (#744) * write mock metadata cache data to mappings _meta Signed-off-by: Sean Kao <seankao@amazon.com> * Enable write to cache by default Signed-off-by: Sean Kao <seankao@amazon.com> * bugfix: _meta.latestId missing when create index Signed-off-by: Sean Kao <seankao@amazon.com> * set and unset config in test suite Signed-off-by: Sean Kao <seankao@amazon.com> * fix: use member flintSparkConf Signed-off-by: Sean Kao <seankao@amazon.com> --------- Signed-off-by: Sean Kao <seankao@amazon.com> * default metadata cache write disabled Signed-off-by: Sean Kao <seankao@amazon.com> * remove string literal "external" in index builder Signed-off-by: Sean Kao <seankao@amazon.com> * track refreshInterval and lastRefreshTime Signed-off-by: Sean Kao <seankao@amazon.com> * add last refresh timestamps to metadata log entry Signed-off-by: Sean Kao <seankao@amazon.com> * update metadata cache test case: should pass Signed-off-by: Sean Kao <seankao@amazon.com> * move to spark package; get refresh interval Signed-off-by: Sean Kao <seankao@amazon.com> * parse refresh interval Signed-off-by: Sean Kao <seankao@amazon.com> * minor syntax fix on FlintSpark.createIndex Signed-off-by: Sean Kao <seankao@amazon.com> * strategize cache writer interface Signed-off-by: Sean Kao <seankao@amazon.com> * update refresh timestamps in FlintSpark Signed-off-by: Sean Kao <seankao@amazon.com> * add test cases Signed-off-by: Sean Kao <seankao@amazon.com> * IT test for refresh timestamp update Signed-off-by: Sean Kao <seankao@amazon.com> * add doc for spark conf Signed-off-by: Sean Kao <seankao@amazon.com> * change mock table name Signed-off-by: Sean Kao <seankao@amazon.com> * add IT test at FlintSpark level Signed-off-by: Sean Kao <seankao@amazon.com> * test with external scheduler Signed-off-by: Sean Kao <seankao@amazon.com> * refactor refreshIndex method; add test for modes Signed-off-by: Sean Kao <seankao@amazon.com> * fix typo Signed-off-by: Sean Kao <seankao@amazon.com> * fix failed test caused by refactoring Signed-off-by: Sean Kao <seankao@amazon.com> * rename method; add comment Signed-off-by: Sean Kao <seankao@amazon.com> --------- Signed-off-by: Sean Kao <seankao@amazon.com> (cherry picked from commit a07f88f)
opensearch-project · Oct 30, 2024 · c3b1da4 · c3b1da4
1 parent f790d35
commit c3b1da4
Show file tree

Hide file tree

Showing 22 changed files with 1,051 additions and 71 deletions.
diff --git a/docs/index.md b/docs/index.md
@@ -549,6 +549,7 @@ In the index mapping, the `_meta` and `properties`field stores meta and schema i
 - `spark.flint.monitor.initialDelaySeconds`: Initial delay in seconds before starting the monitoring task. Default value is 15.
 - `spark.flint.monitor.intervalSeconds`: Interval in seconds for scheduling the monitoring task. Default value is 60.
 - `spark.flint.monitor.maxErrorCount`: Maximum number of consecutive errors allowed before stopping the monitoring task. Default value is 5.
+- `spark.flint.metadataCacheWrite.enabled`: default is false. enable writing metadata to index mappings _meta as read cache for frontend user to access. Do not use in production, this setting will be removed in later version.
 
 #### Data Type Mapping
 

diff --git a/...mmons/src/main/scala/org/opensearch/flint/common/metadata/log/FlintMetadataLogEntry.scala b/...mmons/src/main/scala/org/opensearch/flint/common/metadata/log/FlintMetadataLogEntry.scala
@@ -18,6 +18,10 @@ import org.opensearch.flint.common.metadata.log.FlintMetadataLogEntry.IndexState
  *   log entry id
  * @param state
  *   Flint index state
+ * @param lastRefreshStartTime
+ *   timestamp when last refresh started for manual or external scheduler refresh
+ * @param lastRefreshCompleteTime
+ *   timestamp when last refresh completed for manual or external scheduler refresh
  * @param entryVersion
  *   entry version fields for consistency control
  * @param error
@@ -28,10 +32,12 @@ import org.opensearch.flint.common.metadata.log.FlintMetadataLogEntry.IndexState
 case class FlintMetadataLogEntry(
     id: String,
     /**
-     * This is currently used as streaming job start time. In future, this should represent the
-     * create timestamp of the log entry
+     * This is currently used as streaming job start time for internal scheduler. In future, this
+     * should represent the create timestamp of the log entry
      */
     createTime: Long,
+    lastRefreshStartTime: Long,
+    lastRefreshCompleteTime: Long,
     state: IndexState,
     entryVersion: Map[String, Any],
     error: String,
@@ -40,26 +46,48 @@ case class FlintMetadataLogEntry(
   def this(
       id: String,
       createTime: Long,
+      lastRefreshStartTime: Long,
+      lastRefreshCompleteTime: Long,
       state: IndexState,
       entryVersion: JMap[String, Any],
       error: String,
       properties: JMap[String, Any]) = {
-    this(id, createTime, state, entryVersion.asScala.toMap, error, properties.asScala.toMap)
+    this(
+      id,
+      createTime,
+      lastRefreshStartTime,
+      lastRefreshCompleteTime,
+      state,
+      entryVersion.asScala.toMap,
+      error,
+      properties.asScala.toMap)
   }
 
   def this(
       id: String,
       createTime: Long,
+      lastRefreshStartTime: Long,
+      lastRefreshCompleteTime: Long,
       state: IndexState,
       entryVersion: JMap[String, Any],
       error: String,
       properties: Map[String, Any]) = {
-    this(id, createTime, state, entryVersion.asScala.toMap, error, properties)
+    this(
+      id,
+      createTime,
+      lastRefreshStartTime,
+      lastRefreshCompleteTime,
+      state,
+      entryVersion.asScala.toMap,
+      error,
+      properties)
   }
 }
 
 object FlintMetadataLogEntry {
 
+  val EMPTY_TIMESTAMP = 0L
+
   /**
    * Flint index state enum.
    */

diff --git a/...e/src/main/scala/org/opensearch/flint/core/metadata/log/DefaultOptimisticTransaction.java b/...e/src/main/scala/org/opensearch/flint/core/metadata/log/DefaultOptimisticTransaction.java
@@ -86,6 +86,8 @@ public T commit(Function<FlintMetadataLogEntry, T> operation) {
       initialLog = initialLog.copy(
           initialLog.id(),
           initialLog.createTime(),
+          initialLog.lastRefreshStartTime(),
+          initialLog.lastRefreshCompleteTime(),
           initialLog.state(),
           latest.entryVersion(),
           initialLog.error(),

diff --git a/...ain/scala/org/opensearch/flint/core/storage/FlintMetadataLogEntryOpenSearchConverter.java b/...ain/scala/org/opensearch/flint/core/storage/FlintMetadataLogEntryOpenSearchConverter.java
@@ -101,14 +101,16 @@ public static String toJson(FlintMetadataLogEntry logEntry) throws JsonProcessin
     ObjectMapper mapper = new ObjectMapper();
     ObjectNode json = mapper.createObjectNode();
 
-    json.put("version", "1.0");
+    json.put("version", "1.1");
     json.put("latestId", logEntry.id());
     json.put("type", "flintindexstate");
     json.put("state", logEntry.state().toString());
     json.put("applicationId", applicationId);
     json.put("jobId", jobId);
     json.put("dataSourceName", logEntry.properties().get("dataSourceName").get().toString());
     json.put("jobStartTime", logEntry.createTime());
+    json.put("lastRefreshStartTime", logEntry.lastRefreshStartTime());
+    json.put("lastRefreshCompleteTime", logEntry.lastRefreshCompleteTime());
     json.put("lastUpdateTime", lastUpdateTime);
     json.put("error", logEntry.error());
 
@@ -138,6 +140,8 @@ public static FlintMetadataLogEntry constructLogEntry(
         id,
         /* sourceMap may use Integer or Long even though it's always long in index mapping */
         ((Number) sourceMap.get("jobStartTime")).longValue(),
+        ((Number) sourceMap.get("lastRefreshStartTime")).longValue(),
+        ((Number) sourceMap.get("lastRefreshCompleteTime")).longValue(),
         FlintMetadataLogEntry.IndexState$.MODULE$.from((String) sourceMap.get("state")),
         Map.of("seqNo", seqNo, "primaryTerm", primaryTerm),
         (String) sourceMap.get("error"),

diff --git a/flint-core/src/main/scala/org/opensearch/flint/core/storage/FlintOpenSearchMetadataLog.java b/flint-core/src/main/scala/org/opensearch/flint/core/storage/FlintOpenSearchMetadataLog.java
@@ -132,7 +132,9 @@ public void purge() {
   public FlintMetadataLogEntry emptyLogEntry() {
     return new FlintMetadataLogEntry(
         "",
-        0L,
+        FlintMetadataLogEntry.EMPTY_TIMESTAMP(),
+        FlintMetadataLogEntry.EMPTY_TIMESTAMP(),
+        FlintMetadataLogEntry.EMPTY_TIMESTAMP(),
         FlintMetadataLogEntry.IndexState$.MODULE$.EMPTY(),
         Map.of("seqNo", UNASSIGNED_SEQ_NO, "primaryTerm", UNASSIGNED_PRIMARY_TERM),
         "",
@@ -146,6 +148,8 @@ private FlintMetadataLogEntry createLogEntry(FlintMetadataLogEntry logEntry) {
         logEntry.copy(
             latestId,
             logEntry.createTime(),
+            logEntry.lastRefreshStartTime(),
+            logEntry.lastRefreshCompleteTime(),
             logEntry.state(),
             logEntry.entryVersion(),
             logEntry.error(),
@@ -184,6 +188,8 @@ private FlintMetadataLogEntry writeLogEntry(
       logEntry = new FlintMetadataLogEntry(
           logEntry.id(),
           logEntry.createTime(),
+          logEntry.lastRefreshStartTime(),
+          logEntry.lastRefreshCompleteTime(),
           logEntry.state(),
           Map.of("seqNo", response.getSeqNo(), "primaryTerm", response.getPrimaryTerm()),
           logEntry.error(),

diff --git a/...ala/org/opensearch/flint/core/storage/FlintMetadataLogEntryOpenSearchConverterSuite.scala b/...ala/org/opensearch/flint/core/storage/FlintMetadataLogEntryOpenSearchConverterSuite.scala
@@ -25,6 +25,10 @@ class FlintMetadataLogEntryOpenSearchConverterTest
   val sourceMap = JMap.of(
     "jobStartTime",
     1234567890123L.asInstanceOf[Object],
+    "lastRefreshStartTime",
+    1234567890123L.asInstanceOf[Object],
+    "lastRefreshCompleteTime",
+    1234567890123L.asInstanceOf[Object],
     "state",
     "active".asInstanceOf[Object],
     "dataSourceName",
@@ -36,6 +40,8 @@ class FlintMetadataLogEntryOpenSearchConverterTest
     when(mockLogEntry.id).thenReturn("id")
     when(mockLogEntry.state).thenReturn(FlintMetadataLogEntry.IndexState.ACTIVE)
     when(mockLogEntry.createTime).thenReturn(1234567890123L)
+    when(mockLogEntry.lastRefreshStartTime).thenReturn(1234567890123L)
+    when(mockLogEntry.lastRefreshCompleteTime).thenReturn(1234567890123L)
     when(mockLogEntry.error).thenReturn("")
     when(mockLogEntry.properties).thenReturn(Map("dataSourceName" -> "testDataSource"))
   }
@@ -45,14 +51,16 @@ class FlintMetadataLogEntryOpenSearchConverterTest
     val expectedJsonWithoutLastUpdateTime =
       s"""
          |{
-         |  "version": "1.0",
+         |  "version": "1.1",
          |  "latestId": "id",
          |  "type": "flintindexstate",
          |  "state": "active",
          |  "applicationId": "unknown",
          |  "jobId": "unknown",
          |  "dataSourceName": "testDataSource",
          |  "jobStartTime": 1234567890123,
+         |  "lastRefreshStartTime": 1234567890123,
+         |  "lastRefreshCompleteTime": 1234567890123,
          |  "error": ""
          |}
          |""".stripMargin
@@ -67,15 +75,22 @@ class FlintMetadataLogEntryOpenSearchConverterTest
     logEntry shouldBe a[FlintMetadataLogEntry]
     logEntry.id shouldBe "id"
     logEntry.createTime shouldBe 1234567890123L
+    logEntry.lastRefreshStartTime shouldBe 1234567890123L
+    logEntry.lastRefreshCompleteTime shouldBe 1234567890123L
     logEntry.state shouldBe FlintMetadataLogEntry.IndexState.ACTIVE
     logEntry.error shouldBe ""
     logEntry.properties.get("dataSourceName").get shouldBe "testDataSource"
   }
 
-  it should "construct log entry with integer jobStartTime value" in {
+  it should "construct log entry with integer timestamp value" in {
+    // Use Integer instead of Long for timestamps
     val testSourceMap = JMap.of(
       "jobStartTime",
-      1234567890.asInstanceOf[Object], // Integer instead of Long
+      1234567890.asInstanceOf[Object],
+      "lastRefreshStartTime",
+      1234567890.asInstanceOf[Object],
+      "lastRefreshCompleteTime",
+      1234567890.asInstanceOf[Object],
       "state",
       "active".asInstanceOf[Object],
       "dataSourceName",
@@ -87,6 +102,8 @@ class FlintMetadataLogEntryOpenSearchConverterTest
     logEntry shouldBe a[FlintMetadataLogEntry]
     logEntry.id shouldBe "id"
     logEntry.createTime shouldBe 1234567890
+    logEntry.lastRefreshStartTime shouldBe 1234567890
+    logEntry.lastRefreshCompleteTime shouldBe 1234567890
     logEntry.state shouldBe FlintMetadataLogEntry.IndexState.ACTIVE
     logEntry.error shouldBe ""
     logEntry.properties.get("dataSourceName").get shouldBe "testDataSource"

diff --git a/...t-spark-integration/src/main/scala/org/apache/spark/sql/flint/config/FlintSparkConf.scala b/...t-spark-integration/src/main/scala/org/apache/spark/sql/flint/config/FlintSparkConf.scala
@@ -253,6 +253,10 @@ object FlintSparkConf {
     FlintConfig("spark.metadata.accessAWSCredentialsProvider")
       .doc("AWS credentials provider for metadata access permission")
       .createOptional()
+  val METADATA_CACHE_WRITE = FlintConfig("spark.flint.metadataCacheWrite.enabled")
+    .doc("Enable Flint metadata cache write to Flint index mappings")
+    .createWithDefault("false")
+
   val CUSTOM_SESSION_MANAGER =
     FlintConfig("spark.flint.job.customSessionManager")
       .createOptional()
@@ -309,6 +313,8 @@ case class FlintSparkConf(properties: JMap[String, String]) extends Serializable
 
   def monitorMaxErrorCount(): Int = MONITOR_MAX_ERROR_COUNT.readFrom(reader).toInt
 
+  def isMetadataCacheWriteEnabled: Boolean = METADATA_CACHE_WRITE.readFrom(reader).toBoolean
+
   /**
    * spark.sql.session.timeZone
    */