Rework Iceberg deduplication example (#103)

Author: mbroecheler

.github/workflows/build.yml

@@ -73,7 +73,7 @@ jobs:
           - example: iceberg-dedup
             path: iceberg-data-deduplication
             test_commands: |
-              compile dedup_package.json
+              compile data_deduplication_package.json
 
           - example: iot-sensor
             path: iot-sensor-metrics

.gitignore

@@ -224,3 +224,8 @@ Temporary Items
 
 py-venv
 # End of https://www.toptal.com/developers/gitignore/api/macos
+
+##performance benchmarking
+node_modules/
+iceberg-data/
+sqrl-iceberg-data/

iceberg-data-deduplication/dedup.sqrl …ta-deduplication/data_deduplication.sqrliceberg-data-deduplication/dedup.sqrl renamed to iceberg-data-deduplication/data_deduplication.sqrl iceberg-data-deduplication/dedup.sqrl renamed to iceberg-data-deduplication/data_deduplication.sqrl

@@ -1,14 +1,9 @@
 -- This script runs a deduplication compaction on Iceberg tables
 -- It currently assumes a single primary partition column and a single time bucket partition column based on seconds
 
-IMPORT tables.{{tableName}};
-IMPORT tables.compaction_tracker;
+IMPORT tables.{{tableName}}.*;
+IMPORT tables.compaction_tracker.*;
 IMPORT udfs.read_partition_sizes;
-IMPORT udfs.delete_duplicated_data;
-
---
--- Part 1: Gather partitions to compact
---
 
 _AllPartitions :=
 SELECT partition_map['{{partitionCol}}'] AS partition_id,
@@ -52,41 +47,17 @@ SELECT DISTINCT '{{tableName}}' AS table_name,
                 partition_id,
                 new_rel_percentage,
                 new_abs_percentage,
-                FLOOR((CAST('${DEPLOYMENT_TIMESTAMP}' AS BIGINT) - {{bufferSeconds}}) / {{bucketSeconds}}) * {{bucketSeconds}} AS max_time_bucket,
-                'initialize' AS action,
-                NOW() AS action_time
+                FLOOR((CAST('${DEPLOYMENT_TIMESTAMP}' AS BIGINT) - {{bufferSeconds}}) / {{bucketSeconds}}) * {{bucketSeconds}} AS max_time_bucket
 FROM _PartitionPriority
 WHERE cumulative_total_size <= {{maxTotalPartitionSizeInMB}} * 1024 * 1024
    OR row_num <= 1;
 
-INSERT INTO CompactionTracker SELECT * FROM _PartitionsToCompact;
-
-NEXT_BATCH;
-
---
--- Part 2: Compact gathered partitions
---
-
-_DataToCompact :=
-SELECT DISTINCT table_name,
-                job_id,
-                partition_id,
-                new_rel_percentage,
-                new_abs_percentage,
-                max_time_bucket
-FROM CompactionTracker
-WHERE LOWER(table_name) = LOWER('{{tableName}}')
-  AND job_id = '${DEPLOYMENT_ID}'
-  AND action = 'initialize'
-  AND action_time > NOW() - INTERVAL '4' HOUR; -- last condition is for efficient pruning of iceberg read
-
-_CastDataToCompact := SELECT * FROM _DataToCompact;
-_CastDataToCompact.partition_id := CAST(partition_id AS {{partitionColType}});
+_PartitionsToCompact.partition_id := CAST(partition_id AS {{partitionColType}});
 
 _InputData :=
 SELECT /*+ BROADCAST(c) */ d.*
 FROM {{tableName}} AS d
-JOIN _CastDataToCompact AS c
+JOIN _PartitionsToCompact AS c
   ON d.{{partitionCol}} = c.partition_id
  AND d.{{timeBucketCol}} <= c.max_time_bucket;
 
@@ -98,60 +69,20 @@ INSERT OVERWRITE {{tableName}}
     SELECT * FROM _DistInputData
     -- filter only when "deleteFlagCol" has a value
     {{#deleteFlagCol}}
-    WHERE {{deleteFlagCol}} IS TRUE
+    WHERE {{deleteFlagCol}} IS NOT TRUE
     {{/deleteFlagCol}}
     -- sort by partition column, and optionally any other column(s)
     ORDER BY {{partitionCol}} {{#sortKey}}, {{sortKey}}{{/sortKey}};
 
 _CompactionResult :=
-SELECT table_name,
-       job_id,
-       partition_id,
+SELECT CAST(table_name AS STRING),
+       CAST(job_id AS STRING),
+       CAST(partition_id AS STRING),
        new_rel_percentage,
        new_abs_percentage,
        max_time_bucket,
        'overwrite' AS action,
        NOW() AS action_time
-FROM _DataToCompact;
+FROM _PartitionsToCompact;
 
 INSERT INTO CompactionTracker SELECT * FROM _CompactionResult;
-
-NEXT_BATCH;
-
---
--- Part 3: Delete source partitions that were compacted
---
-
-_DataToDelete :=
-SELECT DISTINCT table_name,
-                job_id,
-                partition_id,
-                new_rel_percentage,
-                new_abs_percentage,
-                max_time_bucket
-FROM CompactionTracker
-WHERE LOWER(table_name) = LOWER('{{tableName}}')
-  AND job_id = '${DEPLOYMENT_ID}'
-  AND action = 'overwrite'
-  AND action_time > NOW() - INTERVAL '4' HOUR;
-
-_DeleteFnResult :=
-SELECT delete_duplicated_data(
-               '{{warehouse}}',
-               '{{catalogType}}',
-               '{{catalogName}}',
-               '{{databaseName}}',
-               MAX(table_name),
-               MAX(max_time_bucket),
-               COLLECT(MAP['{{partitionCol}}', partition_id])) AS res
-FROM _DataToDelete;
-
-_DeleteResult :=
-SELECT
-    d.*,
-    CASE WHEN fn.res = true THEN 'delete' ELSE 'delete_failed' END AS action,
-        NOW() AS action_time
-FROM _DataToDelete AS d
-    CROSS JOIN (SELECT res FROM _DeleteFnResult LIMIT 1) AS fn;
-
-INSERT INTO CompactionTracker SELECT * FROM _DeleteResult;

…rg-data-deduplication/dedup_package.json …lication/data_deduplication_package.jsoniceberg-data-deduplication/dedup_package.json renamed to iceberg-data-deduplication/data_deduplication_package.json iceberg-data-deduplication/dedup_package.json renamed to iceberg-data-deduplication/data_deduplication_package.json

@@ -2,13 +2,13 @@
   "version": "1",
   "enabled-engines": ["flink"],
   "script": {
-    "main": "dedup.sqrl",
+    "main": "data_deduplication.sqrl",
     "config": {
-      "warehouse": "/tmp/duckdb",
+      "warehouse": "sqrl-iceberg-data",
       "catalogType": "hadoop",
       "catalogName": "mycatalog",
       "databaseName": "mydatabase",
-      "tableName": "CdcData",
+      "tableName": "cdc_data",
       "distinctKeyCol": "account_id",
       "partitionCol": "account_id",
       "partitionColType": "bigint",

iceberg-data-deduplication/data_delete.sqrl

@@ -0,0 +1,46 @@
+IMPORT tables.compaction_tracker.*;
+IMPORT udfs.delete_duplicated_data;
+
+_DataToDelete :=
+SELECT table_name,
+       job_id,
+       partition_id,
+       new_rel_percentage,
+       new_abs_percentage,
+       max_time_bucket
+FROM (
+    SELECT table_name,
+           job_id,
+           partition_id,
+           new_rel_percentage,
+           new_abs_percentage,
+           max_time_bucket,
+           action,
+           ROW_NUMBER() OVER (PARTITION BY table_name, partition_id ORDER BY action_time DESC) AS row_num
+    FROM CompactionTracker
+    WHERE action_time < NOW() - INTERVAL '{{commit_wait_sec}}' SECOND AND action_time > NOW() - INTERVAL '24' HOUR -- time buffer to ensure commit happened and stop attempting to delete
+)
+WHERE row_num = 1 AND action <> 'delete'; -- compacted partitions haven't been deleted yet
+
+--TODO: need extra safety: validate that the selected partitions should be deleted by checking that the last partition update time < partition 0 update time
+
+_DeleteFnResult :=
+SELECT delete_duplicated_data(
+               '{{warehouse}}',
+               '{{catalogType}}',
+               '{{catalogName}}',
+               '{{databaseName}}',
+               MAX(table_name),
+               MAX(max_time_bucket),
+               COLLECT(MAP['{{partitionCol}}', partition_id])) AS res
+FROM _DataToDelete;
+
+_DeleteResult :=
+SELECT
+    d.*,
+    CASE WHEN fn.res = true THEN 'delete' ELSE 'delete_failed' END AS action,
+        NOW() AS action_time
+FROM _DataToDelete AS d
+    CROSS JOIN (SELECT res FROM _DeleteFnResult LIMIT 1) AS fn;
+
+INSERT INTO CompactionTracker SELECT * FROM _DeleteResult;

iceberg-data-deduplication/data_delete_package.json

@@ -0,0 +1,22 @@
+{
+  "version": "1",
+  "enabled-engines": ["flink"],
+  "script": {
+    "main": "data_delete.sqrl",
+    "config": {
+      "warehouse": "sqrl-iceberg-data",
+      "catalogType": "hadoop",
+      "catalogName": "mycatalog",
+      "databaseName": "mydatabase",
+      "partitionCol": "account_id",
+      "commit_wait_sec": 1
+    }
+  },
+  "engines": {
+    "flink": {
+      "config": {
+        "execution.runtime-mode": "batch"
+      }
+    }
+  }
+}

iceberg-data-deduplication/datagen.sqrl

@@ -1,4 +1,8 @@
 IMPORT tables.min_source AS _CDC;
-IMPORT tables.CdcData;
+IMPORT tables.{{tableName}}.*;
 
-INSERT INTO CdcData SELECT c.*, EXTRACT(SECOND FROM c.ts) + 1 AS time_bucket FROM _CDC c;
+CountRecords := SELECT COUNT(*) AS num_records FROM _CDC.SrcRecords;
+
+EXPORT CountRecords TO logger.records;
+
+INSERT INTO {{tableName}} SELECT c.*, EXTRACT(SECOND FROM c.ts) + 1 AS time_bucket FROM _CDC.SrcRecords c;

iceberg-data-deduplication/datagen_aws_package.json

@@ -2,7 +2,14 @@
   "version": "1",
   "enabled-engines": ["flink"],
   "script": {
-    "main": "datagen.sqrl"
+    "main": "datagen.sqrl",
+    "config": {
+      "warehouse": "sqrl-iceberg-data",
+      "catalogType": "hadoop",
+      "catalogName": "mycatalog",
+      "databaseName": "mydatabase",
+      "tableName": "cdc_data"
+    }
   },
   "engines": {
     "flink": {

iceberg-data-deduplication/datagen_package.json

@@ -1,4 +1,4 @@
-CREATE TABLE `CdcData` (
+CREATE TABLE `cdc_data` (
     `account_id`  BIGINT NOT NULL,
     `id`          BIGINT NOT NULL,
     `pt`          STRING NOT NULL,
@@ -10,10 +10,10 @@ CREATE TABLE `CdcData` (
 ) PARTITIONED BY (`account_id`, `id`, `time_bucket`)
 WITH (
     'connector' = 'iceberg',
-    'catalog-name' = 'mycatalog',
-    'catalog-database' = 'mydatabase',
-    'catalog-table' = 'cdcdata',
-    'catalog-type' = 'hadoop',
-    'warehouse' = '/tmp/duckdb',
+    'catalog-name' = '{{catalogName}}',
+    'catalog-database' = '{{databaseName}}',
+    'catalog-table' = '{{tableName}}',
+    'catalog-type' = '{{catalogType}}',
+    'warehouse' = '{{warehouse}}',
     'format-version' = '2'
 );