Release notes of Data Cache
Driver Release Note - Internal
Data Cache driver 27.0, from 26.0
| Build | Date | Description |
|---|---|---|
26.0.4 |
22-03-2026 | first build of release 26.0 |
27. |
10-11-2026 | latest build of release 27.0 at the time of writing |
The older end is the first build of release 26.0, not its last, because later builds of 26.0 received ports of changes made for 27.0. The data model of a release is the model that a later build of the same release generated.
- 12 changes
- 1 can break a statement
- 3 groups not available
This Note
Groups
- Data model
- Provider attributes
- Capabilities
- Performance 3
- Bug fixes 5
- Enhancements 2
- Log-on and platform 2
More
What Changed, by Group
Every change is in one group: the group answers the question that you ask before you upgrade.
- Data model
- –
- not available
- Do my statements and views still work?
- Provider attributes
- –
- not available
- Do I have to change my connection string, provider file or settings?
- Capabilities
- –
- not available
- What can the driver do with a table?
- Performance
- 3
- Is it faster, and does it call the platform less?
- Bug fixes
- 5
- Is the defect that I met repaired?
- Enhancements
- 2
- What can I do now that I could not?
- Log-on and platform
- 2
- 1 can break a statement
- Do I have to log on again, or change something because of the platform?
Download Changes as CSV | Data Model of Data Cache
Before You Upgrade
The data models of this step could not be compared, so changes of tables and columns that break a statement are not known. The data model of this driver comes from build 22.1.262-BETA+4120 in the 26.0 workspace and from build 22.1.262-BETA+4120 in the 27.0 workspace; the groups from model files need a model that a build of each release generated, and generating it needs an account that is not at hand.
These changes can stop an existing statement, view, connection string or script from working:
- Oracle Legacy no longer supported as backing database
- Change the backing provider of the Data Cache or Data Replicator from the legacy Oracle provider to the Oracle managed provider.
Top 12 Changes
Changes that can break a statement first, then by importance. Columns are summarised per table in the data model.
- Oracle Legacy no longer supported as backing database
- Log-on and platform
- Breaks an existing statement or script.
- Faster initial load: tables load in parallel, limited by the backing database
- Performance
- Importance 3 of 5.
- Data Replicator error on the backing connection string repaired; events logged in UTC
- Bug fixes
- Importance 3 of 5.
- Cancellation of a running synchronisation and download of messages
- Enhancements
- Importance 3 of 5.
- Removal of forced garbage collection after replication
- Performance
- Importance 2 of 5.
- Smaller and faster cache files by compressing with Zstandard instead of gzip
- Performance
- Importance 2 of 5.
- Fewer null reference errors when metadata or settings are missing
- Bug fixes
- Importance 2 of 5.
- Messages with an action longer than one character no longer block a batch
- Bug fixes
- Importance 2 of 5.
- Refusal of a number of table partition versions per delete below 1
- Bug fixes
- Importance 2 of 5.
- Session heart beat uses the right column for the process owner
- Bug fixes
- Importance 2 of 5.
- Backing database version and service tier are checked by the database drivers
- Log-on and platform
- Importance 2 of 5.
- Short names for three connection attributes and a folder selector for the cache folder
- Enhancements
- Importance 1 of 5.
Data model
Do my statements and views still work? Tables, views, table functions, columns, parameters, keys, code values, packages and procedures that appear, disappear, are renamed or change type.
Not available. The data model of this driver comes from build 22.1.262-BETA+4120 in the 26.0 workspace and from build 22.1.262-BETA+4120 in the 27.0 workspace; the groups from model files need a model that a build of each release generated, and generating it needs an account that is not at hand.
Provider attributes
Do I have to change my connection string, provider file or settings? Attributes of the connection and settings of the driver that appear, disappear or change default or place where they can be set.
Not available. The data model of this driver comes from build 22.1.262-BETA+4120 in the 26.0 workspace and from build 22.1.262-BETA+4120 in the 27.0 workspace; the groups from model files need a model that a build of each release generated, and generating it needs an account that is not at hand.
Capabilities
What can the driver do with a table? Whether a table or table function can be selected from, inserted into, updated and deleted from.
Not available. The data model of this driver comes from build 22.1.262-BETA+4120 in the 26.0 workspace and from build 22.1.262-BETA+4120 in the 27.0 workspace; the groups from model files need a model that a build of each release generated, and generating it needs an account that is not at hand.
Performance
Is it faster, and does it call the platform less? Paging, batching, caching, filters and projections sent to the platform, fewer requests, less memory.
- 3 changes
- commits
Faster initial load: tables load in parallel, limited by the backing database
The initial load now loads several tables at the same time, each on its own thread and in random order, up to the limit of the backing database and at most 32. A setting for the maximum number of parallel refreshes remains available.
- importance 3
- confidence medium
Removal of forced garbage collection after replication
The cache no longer forces a full memory clean-up after replication. The three settings that controlled this (memory load threshold, replication interval, minimum interval) are no longer offered.
- importance 2
- confidence medium
Smaller and faster cache files by compressing with Zstandard instead of gzip
Result sets stored in the disk cache are compressed with a faster method that gives smaller files. The cache is rebuilt where needed.
- importance 2
- confidence medium
Bug fixes
Is the defect that I met repaired? Wrong or missing rows, wrong values, crashes and wrong messages, repaired in a release line that had the defect.
- 5 changes
- commits
Data Replicator error on the backing connection string repaired; events logged in UTC
The anonymised form of the backing connection string was reloaded before the connection was opened, which led to an error on the Data Replicator. It is now set after opening. Start times of events in the event log are now written in UTC instead of local time.
- importance 3
- confidence medium
Fewer null reference errors when metadata or settings are missing
Several places that assumed that settings, table function parameters or version information exist now check for this first and fail with a normal error or skip the item.
- importance 2
- confidence medium
Messages with an action longer than one character no longer block a batch
The action of incoming messages is stored in a column of one character. The value is now converted to that character and left empty when it is missing or longer, instead of failing the whole batch.
- importance 2
- confidence medium
Refusal of a number of table partition versions per delete below 1
The setting for the number of table partition versions per delete of facts is rejected when it is below 1, with a message and a hint. A value of 0 left the delete loop without progress.
- importance 2
- confidence medium
Session heart beat uses the right column for the process owner
The column for the process owner was addressed under a name without underscore, which does not match the repository column process_owner. The session registration now uses the correct name.
- importance 2
- confidence medium
Enhancements
What can I do now that I could not? New functions and options that are not a table or a column.
- 2 changes
- commits
Cancellation of a running synchronisation and download of messages
Long running synchronisation, the download of incoming messages, the delete of forwarded messages and compression of facts now check for cancellation, so a statement can be stopped.
- importance 3
- confidence medium
Short names for three connection attributes and a folder selector for the cache folder
The attributes for the maximum length of logical view names and column names and the purge interval of the event log have shorter names. The old names remain accepted. The cache folder attribute is marked as a directory.
- importance 1
- confidence high
Log-on and platform
Do I have to log on again, or change something because of the platform? Log-on and token handling, scopes and permissions, a new version of the platform API, retired endpoints and changed limits.
- 2 changes
- commits
Oracle Legacy no longer supported as backing database
The Data Cache cannot use the legacy Oracle provider as backing database any more. The Oracle provider for managed code is supported.
What to Do
Change the backing provider of the Data Cache or Data Replicator from the legacy Oracle provider to the Oracle managed provider.
- can break a statement
- importance 3
- confidence medium
Backing database version and service tier are checked by the database drivers
The Data Cache asks the driver of the backing database whether it is supported as host of the cache, including its minimum version and for Azure its service tier.
- importance 2
- confidence low
More Information
- Data model documentation of the driver
- Questions and answers on the forum
- Documentation of the platform
- All release notes
How This Note Was Made
- 166 lines removed from DataCacheProvider.cs and CheckSupportedAsDataCacheBackingDatabase added in AnsiSqlProviderBase, SqlServer, Oracle, MySql, PostgreSql providers (other directories). Subject says "Add checks"; whether thresholds are stricter than 26.0 was not compared.
- Commit items were classified by one reader; the model changes are derived and need no reading.
- Commits: the commits of the driver directory in the range that have no equivalent patch in the older state, without merges and version stamps; each was classified by reading its message and change.
- Model: the NDJSON files of data-models/database of the workspace of each release are compared record by record; only fields that both models have are compared, and texts are counted, not listed.
- A model generated from a later build of the same release (for example 26.0.340 for 26.0) stands for the release.
- 17 links between a change and a source commit are recorded.
- 4 exclusions are recorded in the audit of the note.