. . . . . . . . . . 55 VStream Skew Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 VStream Skew Detection . . . . . . . . . . . References: Streaming Vitess at Bolt VStream Skew Minimization description: Aligning streams from different shards in the VStream API VStream Skew Detection Motivation When the VStream API is streaming the insert from the second. This would result in the row going “invisible” for the duration of the skew. This can affect user experience in applications where the VStream events are used to refresh UI,

. . . . . . . . . . 52 VStream Skew Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 VStream Skew Detection . . . . . . . . . . . References: Streaming Vitess at Bolt VStream Skew Minimization description: Aligning streams from different shards in the VStream API 52 VStream Skew Detection Motivation When the VStream API is streaming the insert from the second. This would result in the row going “invisible” for the duration of the skew. This can affect user experience in applications where the VStream events are used to refresh UI,

. . . . . . . . . . 77 VStream Skew Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 VStream Skew Detection . . . . . . . . . . . References: Streaming Vitess at Bolt VStream Skew Minimization description: Aligning streams from different shards in the VStream API 77 VStream Skew Detection Motivation When the VStream API is streaming the insert from the second. This would result in the row going “invisible” for the duration of the skew. This can affect user experience in applications where the VStream events are used to refresh UI,

vttablet sends its current time along with every event. This allows the target to correct for clock skew while estimating replication lag. Additionally, the source starts sending heartbeats if there is nothing vttablet sends its current time along with every event. This allows the target to correct for clock skew while estimating replication lag. Additionally, the source starts sending heartbeats if there is nothing specific table. If the increase is across the board, then it is likely a general overload. The mitigation is to increase provisioning or address the root cause of the overload if it was not expected.

Percolator model. Despite promising snap- shot isolation, TiDB 2.1.7 through 3.0.0-beta.1-40 allowed read skew and lost updates by default, thanks to two auto-retry mechanisms which blindly re-applied updates TiDB also supports a select ... for update statement which mostly, but not entirely, prevents write skew. We found a minor race condition in table creation, reduced durability & fault tolerance in fresh dependence on CLOCK_MONOTONIC_RAW for safety, but we have not yet observed anomalies due to clock skew, possibly due to tooling limitations. PingCAP has published a companion blog post to this report.

Orientation 262 Compression 263 Distributions 263 Local (Co-located) Joins 264 Data Skew 264 Processing Skew 265 VMware Greenplum v6.17 Documentation VMware, Inc. 11 Partitioning 267 Number of Checking for Data Distribution Skew 539 Viewing a Table's Distribution Key 539 Viewing Data Distribution 540 Checking for Query Processing Skew 540 Avoiding an Extreme Skew Warning 541 Viewing Metadata Benchmarks 928 Distribution and Skew 928 Distribution and Skew 0 Local (Co-located) Joins 929 Data Skew 929 Considerations for Replicated Tables 930 Processing Skew 930 Common Causes of Performance

Orientation 329 Compression 330 Distributions 330 Local (Co-located) Joins 331 Data Skew 331 Processing Skew 332 Partitioning 332 Number of Partition and Columnar Storage Files 333 Indexes 334 Checking for Data Distribution Skew 607 Viewing a Table's Distribution Key 607 Viewing Data Distribution 608 Checking for Query Processing Skew 608 Avoiding an Extreme Skew Warning 609 VMware Greenplum Benchmarks 996 Distribution and Skew 996 Distribution and Skew 0 Local (Co-located) Joins 997 Data Skew 997 Considerations for Replicated Tables 998 Processing Skew 998 Common Causes of Performance

Orientation 338 Compression 339 Distributions 339 Local (Co-located) Joins 340 Data Skew 340 Processing Skew 341 Partitioning 341 Number of Partition and Columnar Storage Files 342 Indexes 343 Checking for Data Distribution Skew 616 Viewing a Table's Distribution Key 616 Viewing Data Distribution 617 Checking for Query Processing Skew 617 Avoiding an Extreme Skew Warning 617 VMware Greenplum Benchmarks 1005 Distribution and Skew 1005 Distribution and Skew 0 Local (Co-located) Joins 1006 Data Skew 1006 Considerations for Replicated Tables 1007 Processing Skew 1007 Common Causes of Performance

Checking for Data Distribution Skew 1143 Viewing a Table's Distribution Key 1143 Viewing Data Distribution 1143 Checking for Query Processing Skew 1144 Avoiding an Extreme Skew Warning 1144 Checking for Performance Benchmarks 1174 Distribution and Skew 1174 Local (Co-located) Joins 1175 Data Skew 1175 Considerations for Replicated Tables 1176 Processing Skew 1176 Common Causes of Performance Issues Orientation 1243 Compression 1244 Distributions 1244 Local (Co-located) Joins 1245 Data Skew 1245 Processing Skew 1246 Partitioning 1246 Number of Partition and Columnar Storage Files 1247 Indexes 1247

Checking for Data Distribution Skew 1111 Viewing a Table's Distribution Key 1111 Viewing Data Distribution 1111 Checking for Query Processing Skew 1112 Avoiding an Extreme Skew Warning 1112 Checking for Documentation VMware, Inc. 53 Distribution and Skew 1142 Local (Co-located) Joins 1143 Data Skew 1143 Considerations for Replicated Tables 1144 Processing Skew 1144 Common Causes of Performance Issues Orientation 1211 Compression 1212 Distributions 1212 Local (Co-located) Joins 1213 Data Skew 1213 Processing Skew 1214 Partitioning 1214 Number of Partition and Columnar Storage Files 1215 Indexes 1215