Cloud Feature Wait Types in Boxed SQL Server
A wait type name can look as if it came from a cloud feature, even when I am looking at a boxed SQL Server instance. The name is a clue, but it does not establish a feature relationship. For this post, I checked three things. Does the name exist in the product? Does Microsoft link that name to a feature? Did it collect wait statistics on my test instance?
This post is part of the Undocumented Changes in SQL Server 2022 and 2025 series. The contents page lists every part in order.
Earlier posts examined SQL Server 2022 wait types, SQL Server 2025 columnstore transcoder wait types, and the SQL Server 2025 sp_dw and Trident surface. I am narrowing the search to names that contain SYNAPSE, SYNAPSELINK, TRIDENT, DW_, or PARQUET.
The inventory
I found 58 distinct wait type names across the four inventories. SQL Server 2019 had 2, SQL Server 2022 had 34, SQL Server 2025 RTM had 58, and SQL Server 2025 CU9 had the same 58. RTM means the initial release build. CU means cumulative update.
| Wait type | 2019 | 2022 | 2025 RTM | 2025 CU9 |
|---|---|---|---|---|
DW_ADD_DATABASE |
Y | Y | ||
DW_ATTACHMENT_TASK_LIST |
Y | Y | ||
DW_ATTACH_DBS_AND_UPDATE_SECRETS |
Y | Y | Y | |
DW_DB |
Y | Y | Y | |
DW_DB_ATTACHMENT |
Y | Y | ||
DW_INGESTION_CHAMELEON_PREEMPTIVE_CALL |
Y | Y | Y | |
DW_INIT_WORKSPACES |
Y | Y | ||
DW_INIT_WS_INFO_VIA_HTTP_CALLBACK, |
Y | Y | ||
DW_LEASE_RENEW_EVENT |
Y | Y | Y | |
DW_LEASE_RENEW_LOCK |
Y | Y | Y | |
DW_PHYSICAL_METADATA_HYDRATION |
Y | Y | Y | |
DW_RECREATE_WORKSPACES, |
Y | Y | ||
DW_RSCENTRY |
Y | Y | Y | |
DW_SHARED_DB_ATTACHMENT_LIST |
Y | Y | ||
DW_SYNCHRONOUS_ACTIVATION |
Y | Y | ||
DW_TABLE_METRICS_MAP |
Y | Y | ||
DW_TRAN_CREATE |
Y | Y | Y | Y |
DW_WORKSPACE_GROUP_UNITTEST |
Y | Y | ||
DW_WORKSPACE_MAP_MUTEX |
Y | Y | ||
DW_WS_DB_LIST |
Y | Y | Y | |
PARQUET_INDEX_BUILD_INFO_CELLID_SYNC |
Y | Y | ||
PARQUET_INDEX_BUILD_MANIFEST_SYNC |
Y | Y | Y | |
PREEMPTIVE_SYNAPSESTREAMING_HTTP_EVENT_WAIT |
Y | Y | Y | |
PWAIT_DW_DELTA_TABLE_LOG_DESTRUCTION |
Y | Y | ||
PWAIT_SYNAPSE_LINK_CAPTURE |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_COMMIT |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_DATA_EXPORT_SESSION |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_DB_CLEANUP |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_END_HISTORY_SESSION |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_GET_CURRENT_DB_LSN |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_GET_DB_LIST, |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_GET_TABLE_HASHTABLE, |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_LZN_API_CALL |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_MEM_CAP_THROTTLE |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_POPULATE_METADATA |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_PUBLISH |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_PUBLISHER_DONE |
Y | Y | Y | |
PWAIT_SYNAPSE_LINK_UPDATE_TABLE_STATUS |
Y | Y | Y | |
PWAIT_TRIDENT_LINK_NOTIFICATION_WORKITEM_QUEUE |
Y | Y | ||
SYNAPSELINK_CAPTURE_JOBTASK_ACCESS |
Y | Y | Y | |
SYNAPSELINK_COMMIT_JOBTASK_ACCESS |
Y | Y | Y | |
SYNAPSELINK_FAILBATCH_ACCESS |
Y | Y | Y | |
SYNAPSELINK_PUBLISH_JOBTASK_ACCESS |
Y | Y | Y | |
SYNAPSELINK_SNAPSHOT_JOBTASK_ACCESS |
Y | Y | Y | |
TRIDENTLINK_COMMIT_NOTIFICATION_JOBTASK_ACCESS |
Y | Y | Y | |
TRIDENT_COMMIT_NOTIFICATION |
Y | Y | ||
TRIDENT_DW_MWC_HTTP_CLIENT_INIT_MUTEX |
Y | Y | ||
TRIDENT_DW_REPORT_USAGE_BUFFER_INIT_MUTEX |
Y | Y | ||
TRIDENT_DW_REPORT_USAGE_BUFFER_WRITE_MUTEX |
Y | Y | ||
TRIDENT_DW_REPORT_USAGE_PERIODIC_TASK |
Y | Y | ||
TRIDENT_LINK_GET_LZN_CURRENT_TABLE_VERSION |
Y | Y | ||
TRIDENT_ONELAKE_CLIENT_PROXY |
Y | Y | ||
TRIDENT_ONELAKE_ENDPOINT |
Y | Y | Y | |
TRIDENT_ONELAKE_REGEXP_MATCH |
Y | Y | ||
TRIDENT_OPMETRICS_RWLOCK |
Y | Y | ||
WAIT_DW_TX_EXTERNALIZATION_IO_COMPLETION |
Y | Y | Y | |
WAIT_MWC_DW_EXTENSIBILITY_TOKEN_CALLBACK_INITLOCK |
Y | Y | ||
WAIT_XCS_PARQUET_SEGMENT_IO |
Y | Y | Y | Y |
Most of the Synapse Link names arrived in SQL Server 2022. The DW_, TRIDENT_*, and extra PARQUET_* names are mostly SQL Server 2025 additions. Two names were already present in the 2019 inventory: DW_TRAN_CREATE and WAIT_XCS_PARQUET_SEGMENT_IO. Both remained present through SQL Server 2025 CU9, so I did not treat them as new cloud-feature additions.
What Microsoft says about the feature names
Microsoft documents Azure Synapse Link for SQL as a way to get near real-time analytics over operational data. The supported sources are Azure SQL Database and SQL Server 2022. The same overview now tells readers to use Fabric Mirroring instead of Azure Synapse Link.[1] The SQL Server discontinued-functionality page is more direct for the boxed product. It says Synapse Link is discontinued in SQL Server 2025, and says to use Mirroring in Fabric instead.[2]
The inventory therefore contains Synapse Link wait type names in SQL Server 2025. SQL Server 2022 adds the Synapse Link wait type names. SQL Server 2025 still has those names in the inventory, even though Microsoft lists Synapse Link as discontinued for SQL Server 2025. The inventory does not show whether the feature works. It only shows that the wait type names remain in sys.dm_os_wait_stats.
Fabric Mirroring is the replacement Microsoft points to. The Fabric mirroring page says mirroring can continuously replicate SQL Server databases into Fabric OneLake. It also says SQL Server 2025 mirroring uses the Fabric mirroring change feed feature. Mirroring for SQL Server 2016 to 2022 uses Change Data Capture (CDC).[3]
The DW_ and TRIDENT_* names need more care. Microsoft documents sys.sp_dw_refresh_ext_table, but its Applies to list is Fabric SQL analytics endpoint, mirrored databases in Fabric, and SQL database in Fabric. It is not documented as a SQL Server Database Engine procedure.[4] Microsoft also documents the Fabric SQL analytics endpoint as the endpoint that keeps SQL metadata synchronized with Delta tables in OneLake.[5]
Those pages associate sp_dw_refresh_ext_table with Fabric. They do not identify the TRIDENT_* wait type names as a SQL Server Database Engine feature. They also do not define TRIDENT. I found public Microsoft references to Trident, including SQL Server error text, but not a public definition of the term or of these wait types. I am leaving TRIDENT_* as a name-based clue, not a documented feature link.
The PARQUET names are also name-based clues unless a page names the exact wait type. Microsoft documents PolyBase support for Parquet. The PolyBase overview says SQL Server 2025 can use OPENROWSET, CREATE EXTERNAL TABLE, and CREATE EXTERNAL TABLE AS SELECT with several external-data paths. The listed paths are Parquet, Delta, Azure Blob Storage, Azure Data Lake Storage, and S3-compatible object storage. That support no longer requires installing the PolyBase Query Service for External Data.[6] That supports Parquet as a boxed SQL Server feature area. It does not document PARQUET_INDEX_BUILD_INFO_CELLID_SYNC, PARQUET_INDEX_BUILD_MANIFEST_SYNC, or WAIT_XCS_PARQUET_SEGMENT_IO. Those names remain undocumented by the checks I ran, so I label the Parquet feature link as an inference from the name.
Documentation status for the wait types
I checked the parsed sys.dm_os_wait_stats Learn table first. All 58 wait type names were absent from that table. I also searched the SQL Server, Azure, and Fabric documentation source repositories with whole-word, fixed-string git grep, then repeated the search case-insensitively. I found no exact wait type name matches.
The names alone are not sufficient for troubleshooting guidance. PWAIT_SYNAPSE_LINK_CAPTURE looks related to Synapse Link because of its name and timing. Microsoft documents the feature, and Microsoft documents that the feature is discontinued in SQL Server 2025. But I did not find a public description of the wait type itself.
The same limitation applies to TRIDENT_ONELAKE_CLIENT_PROXY, TRIDENT_DW_REPORT_USAGE_PERIODIC_TASK, and PWAIT_TRIDENT_LINK_NOTIFICATION_WORKITEM_QUEUE. The names contain OneLake, DW, and Trident. Microsoft documents OneLake and Fabric SQL endpoints elsewhere. I did not find Microsoft documentation for these wait types in the sources I checked. I searched exact names, not only broad feature terms, because feature pages can exist without documenting the wait type names.
What accumulated on the 2025 instance
I ran one read-only query against the 2025 instance. It checks the same name patterns in sys.dm_os_wait_stats, the dynamic management view (DMV) that stores cumulative wait statistics.
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
SET NOCOUNT ON; SELECT [wait_type] , [waiting_tasks_count] , [wait_time_ms] , [signal_wait_time_ms] , [max_wait_time_ms] FROM sys.dm_os_wait_stats WHERE [wait_type] LIKE N'%SYNAPSE%' OR [wait_type] LIKE N'%SYNAPSELINK%' OR [wait_type] LIKE N'%TRIDENT%' OR [wait_type] LIKE N'%DW[_]%' OR [wait_type] LIKE N'%PARQUET%' ORDER BY [wait_type]; |
The query returned all 58 names on the SQL Server 2025 CU9 test instance. Every row had waiting_tasks_count = 0, wait_time_ms = 0, signal_wait_time_ms = 0, and max_wait_time_ms = 0.
This result applies only to the idle test instance. It says that the instance had not accumulated completed waits for these names. It does not prove that the names never accumulate. It also does not prove that the related feature paths are inactive on every boxed SQL Server instance.
How I would read this in monitoring
If one of these wait type names appears in top waits, I would start with the exact feature evidence, not the name alone. For Synapse Link names, the feature is documented for SQL Server 2022 and discontinued for SQL Server 2025. For Fabric Mirroring in SQL Server 2025, Microsoft documents a different feature path that uses the Fabric mirroring change feed. For Trident, I did not find a public expansion or feature definition. For Parquet, SQL Server supports Parquet through PolyBase, but the wait types themselves are not documented.
The boxed product can carry wait type names that point toward cloud features, retired features, or implementation details that Microsoft has not documented publicly. This inventory shows that the names are present and that the documentation does not describe them. My idle SQL Server 2025 test instance had zero accumulated waits for all 58.
Have you seen one of these wait types collect counts on a real workload, or do you know of documentation my searches missed? Tell me on Bluesky or LinkedIn, and I’ll update the notes.
References
- What is Azure Synapse Link for SQL? - Microsoft Learn. Says Azure Synapse Link for SQL supports SQL Server 2022 and now recommends Fabric Mirroring instead. ↩
- Discontinued Database Engine Functionality - Microsoft Learn. Says Synapse Link is discontinued in SQL Server 2025 and recommends Mirroring in Fabric instead. ↩
- Microsoft Fabric Mirrored Databases From SQL Server - Microsoft Learn. Documents Fabric Mirroring for SQL Server and says SQL Server 2025 mirroring uses the Fabric mirroring change feed feature. ↩
- sys.sp_dw_refresh_ext_table (Transact-SQL) - Microsoft Learn. Documents this
sp_dw_*procedure for Fabric SQL analytics endpoint, mirrored databases in Fabric, and SQL database in Fabric. ↩ - SQL Analytics Endpoint Metadata Sync - Microsoft Learn. Describes SQL analytics endpoint metadata sync for lakehouse, warehouse, database, and mirrored database items in Fabric. ↩
- PolyBase overview for SQL Server - Microsoft Learn. Documents SQL Server 2025 native support for Parquet, Delta, Azure Blob Storage, Azure Data Lake Storage, and S3-compatible object storage paths without the PolyBase Query Service for External Data. ↩