CDM Library Release Notes¶
The first public release of CDM was version 184.108.40.206, release date April 15, 2018.
Versions with the same major and minor release number use the same protocol between CDM client and service, and can be interchanged freely. Higher release numbers for a given major and minor release number contain bug fixes and potentially new features that do not break the client-server protocol. The on-demand cloud CDM service, will always start a CDM service instance of the latest available release for a given major and minor release number.
New Features and Bug Fixes¶
- 220.127.116.11 [09-02-2021]
cdm::CommitElementInCategory()could create negative label numbers in the CDM database, when additional elements were created in a set next to the one offered as an argument to the function.
Added retry capability for cloud CDM service, which may time out and terminate in between obtaining the service URL and the actual connection attempt.
- 18.104.22.168 [20-11-2020]
Snapshot updating mechanism could end up in an infinite loop performing a check every millisecond.
Reduce autotermination period by 1 minute.
- 22.214.171.124 [11-09-2020]
cdm::RevisionBranchwould result in dense execution, taking excessively long for a large number of revisions.
- 126.96.36.199 [09-09-2020]
cdm::GetValuesLog()could produce no values if some domain elements in the log values domain or range were not present in the current contents of the corresponding domain sets. Such tuples are now skipped, and the number of skipped values is reported in the log file.
- 188.8.131.52 [03-09-2020]
Server-side lock was being held for too long, causing a dead-lock when multiple
cdm::CreateSnapshot()requests were fired at the same time.
- 184.108.40.206 [31-08-2020]
Accessing multiple CDM application databases within a single database server would lead to a separate collection of database connections being used for every application database. All access to CDM application databases within a single database server will now use a shared connection pool, and connections in the pool will be automatically garbage collected after 15 minutes of inactivity.
- 220.127.116.11 [27-08-2020]
Some definitions of sets in the CDM library gave syntax and semantic errors in the cloud, preventing CMD apps from being published.
The thread for automatically updating snapshots could crash the CDM service when a database connection was misconfigured.
cdm::DetermineCategoryOrderdid not fully compute all category dependencies. Because this makes the check for cross-dependencies stricter, in rare cases this might lead to a re-ordering of cross-dependent categories and a potential change in the loading order of data if a model actually has dependency problems with its CDM categories.
- 18.104.22.168 [17-08-2020]
Subsets were not filtered during checkout to only pass the non-empty elements.
- 22.214.171.124 [12-08-2020]
Re-committing unmapped labels when a client category was not up-to-date, could cause a crash in the CDM server.
Addded new function to fill
cdm::Categorieswithout actually having to call
- 126.96.36.199 [05-08-2020]
cdm::CreateBranch()will now automatically update the set
cdm::Brancheswith the new branch information.
- 188.8.131.52 [21-07-2020]
Addresses a performance degradation in computing differences between current and committed data.
cdm::DeleteDependentBranches()could delete branches originating after the given end revision.
cdm::PullChanges()could fail to use cached commits when called from a commit notification if two categories were committed intermittently, leading to increased pull times in the presence of multiple clients auto-pulling the changes.
- 184.108.40.206 [16-07-2020]
Changing 0.0 to zero would not be detected by CDM because of the semantics of numerical
<>operator in AIMMS.
- 220.127.116.11 [14-07-2020]
Unitialized local variable could cause crash on Linux.
- 18.104.22.168 [01-07-2020]
Fixed missing symbol in libcdm.so on Linux
- 22.214.171.124 [24-06-2020]
Changesets are now compressed during transport to reduce transmission time and in database cache to reduce stored snapshot size.
Introduced separate function
cdm::CreateSnapshot()to create a cached snapshot asynchronously and completely server-side.
Removed the optional
cdm::RetireBranchData()has been implemented in a totally different manner because a fix to the previous implementation fundamentally prevented it from working for SQLServer-backed CDM instances.
Stopped supporting VC120-based AIMMS versions.
NB. Because the wire and storage format for snapshots changed, all cached snapshots stored in the CDM database will be deleted. Also, the function prototypes for creating snapshots and retiring branch data are changed. If you used these functionalities before, you should update your model.
- 126.96.36.199 [10-03-2020]
cdm::DataChangeProcedurepass on exception only on last retry.
- 188.8.131.52 [25-02-2020]
CDM runtime identifiers for identifiers with defaults and a derived unit, would inadvertently get a default in the base unit, leading to unnecessary commits to the CDM database.
Identifier-specific commit cardinalities could fail the cdm::GetRevisions function for identifiers that no longer exist in the model
The CDM runtime could fail when retrieving branch data for branch- and revision-related identifiers in the CDM library with different internal AIMMS storage types.
- 184.108.40.206 [17-02-2020]
Listen to incoming commit notifications in default callback
cdm::DataChangeProcedureto minimize the chance for
cdm::CommitChangesto fail for auto-commit categories.
- 220.127.116.11 [14-02-2020]
Fixed membership check for element parameters into root sets.
Deleted root set elements would not be deleted properly from other sessions in all circumstances.
Re-order changeset handling such that all changesets are retrieved prior to handling all element space changes of all changesets prior to handling all data changes of all changesets in order to prevent root set mismatches when reading multi-dimensional data from a snapshot in some category associated with a root set from another category where the element was deleted during a revision after the snapshot revision.
Make rollback more robust against element parameters holding inactive values.
When committing root sets adapt labelmembership of element space.
Check for incoming notifications after waiting for data changes to allow notifications to be handled prior to auto-committing.
Function to retrieve branchname would actually try to find branch name in databases set.
Elements of defined root sets would not always be committed immediately the first commit after database creation.
- 18.104.22.168 [12-02-2020]
Data changes for identifiers in some category associated with set elements added and removed to a root set contained in another category in a revision range loaded after a cached snapshot would lead to a runtime error, because such set elements would not be contained in this root set when loading the data. Data changes for such elements are now filtered out when loading the data in the AIMMS client.
In rare occassions, CDM could try to retrieve the element name of set elements that were registered as being added at one time, but removed from the model later on, leading to faulty element names. Element names are now registered when the corresponding newly added elements are discovered by CDM.
NB. This fix required a change in the format of the changesets sent over the wire, which is also the format of the cached snapshots in the CDM database. Consequently, any existing old-format snapshots stored in the CDM database will be deleted on first load, and should be re-created from within the CDM-enabled application.
- 22.214.171.124 [05-02-2020]
Labels added prior to a snapshot revision, but then removed from the set in the snapshot revision, could lead to client-side data loss when such a label was re-added as part of a revision range passed to the client during a checkout based on a cached snapshot.
- 126.96.36.199 [30-01-2020]
Having predeclared identifiers in
cdm::AllCDMIdentifierswould make the call to
AttributeToStringfail PRO solver sessions.
cdm::IdentifierOrderOverrideto CDM library to allow manually setting identifier order for category identifiers set via
- 188.8.131.52 [29-01-2020]
Pull changesets being appended to checkout snapshots could represent revision ranges that add root set elements with associated data, and subsequently delete such elements, leading to partially failed checkouts because of inactive data when handling the changeset.
Domain errors when pulling in changes would only appear in log files and not in client session.
- 184.108.40.206 [22-01-2020]
Predeclared identifiers could not be part of any category.
- 220.127.116.11 [21-01-2020]
When contents of root sets was added thru multiple change sets during checkout (e.g. when using cached checkout snapshots), the root set would only contain the elements added during the last change set.
Recompile CDM runtime library before calling action procedures to prevent compile errors due to edit actions in other runtime libraries such as the WebUI runtime library.
- 18.104.22.168 [16-01-2020]
Data manipulations involving shadow identifiers when committing, checking out and pulling changes, are now running faster by executing them in a procedure in the CDM runtime library, instead of retrieving, comparing and setting all data thru the AIMMS API.
- 22.214.171.124 [11-01-2020]
cdm::IdentifierCategoryOverrideto CDM library to allow adding identifiers from read-only libraries to categories
Up until release 126.96.36.199, set membership for newly added labels to any (non-integer) root set in your model was never set explicitly, but was always implicitly set server-side when such labels were presented to the CDM service. In support of the commit changeset caching feature introduced in CDM release 1.15, set membership is now always required to be set explicitly, but explicitly setting set membership is only possible if the root set is actually contained in some category in your CDM setup. However, for any root set that is part of read-only libraries of your model, adding it to a category was impossible because it was impossible to add the
cdm::categoryannotation. Through the identifier
cdm::IdentifierCategoryOverride, you now have the ability to add such root sets to a CDM category.
- 188.8.131.52 [10-01-2020]
Terminating the cache update thread would crash AIMMS developer when closing a project running an embedded CDM service
- 184.108.40.206 [08-01-2020]
Failed commit could lead to labels to be translated to non-existent label numbers in subsequent commits
Label numbers erroneously ending up with an empty label name in the database could confuse the corresponding set in model and lead to an execution error; such labels are now skipped
Fix a potential commit error when committing to a newly created database a label that was added as a default to an element parameter
Commit changesets are now cached, allowing other clients pulling the same changeset due to a commit notification to retrieve it without any database access, leading to a drastic reduction in database load and pull timings
Checkout snapshots for a specific category-branch combination can now be cached, with a specified interval for the cached snapshot to be updated by the server. Checkout requests on the same category-branch combination will now look for a cached snapshot, and combine this with a pull request from the cached snapshot to the head of the branch to produce the requests checkout. When snapshot caching is enabled, this will lead to drastically reduced checkout times.
For CDM backends backed by a MySQL database, you may need to increase the value of the MySQL option
max_allowed_packetfor categories containing a lot of data. If packet size is not big enough to contain the entire snapshot, the connection to the database will be lost when the CDM service tries to store the snapshot.
- 220.127.116.11 [24-10-2019]
Left-over temporary tables are now removed at service startup
- 18.104.22.168 [14-10-2019]
Checkout of a simple integer subset with large amount of both element additions and deletions could lead to crash
- 22.214.171.124 [04-10-2019]
cdm::DefaultCommitInfoNotification()to allow strictly sequential pulling per commit per category in order to maintain proper cross-category root set - subset relationships in special cases.
- 126.96.36.199 [03-10-2019]
Changes in multi-dimensional identifiers due to data becoming inactive due to elements being removed from domain sets that were true subsets were committed on the first real change to such identifiers. Changes due to data becoming inactive are now never committed regardless of whether the domain sets are root set or subsets.
Yet unhandled data change events could cause the function
- 188.8.131.52 [27-09-2019]
Selected sensible default and alternative filter strategies for all supported databases.
Added commit timeout next to call timeout argument in
cdm::ConnectToApplicationDatabase(), and lowered default call timeout.
Suppressed commit dialog that appeared when commits lasted at least 60 seconds in the WinUI by default.
Added customizable notification and datachange procedures to
cdm::CreateCategories()call as well
Introduced state machine for correctly keeping CDM identifier state in all use cases
Merging in external data could lead to AIMMS errors in certain situations
Commit notifications could be held back by the CDM DLL, causing certain revisions of some categories not to be updated as much as they could by the default commit notification procedure. All commit notifications are now forwarded to the specified commit notification procedure in the model.
cdm::WaitForCommitNotifications()function, to allow the model to wait for and execute commit notifications synchronously prior to e.g. committing category changes to minimize the chance of failed commits due to running behind compared to the CDM server.
- 184.108.40.206 [29-08-2019]
Index columns of multidimensional identifier tables were not declared as
Added option to database configuration file to convert schema and table names to lower case.
- 220.127.116.11 [27-08-2019]
Improved code to implement CDM schema update CDM-2019-06-01 to prevent empty column names for redefined tables.
- 18.104.22.168 [21-08-2019]
CDM schema update CDM-2019-06-01 could leave upgraded CDM databases with wrong value column names
Introduced runtime parameter to allow for alternative filtering strategy that works more performant for a low active/total ratio of domain set elements during checkout.
- 22.214.171.124 [20-08-2019]
Failed data pull would rollback local changes instead of clearing delta-in identifiers.
cdm::ConnectToCategory()could be called multiple times, leading to multiple commit notifications being fired to single client.
Category-dependent notification and datachange procedures communicated when calling
cdm::ConnectToApplicationDB()can now be set via element parameters
- 126.96.36.199 [31-07-2019]
Translation vectors for set elements could be resized too small when extending sets, leading to potential data loss
- 188.8.131.52 [18-07-2019]
Multiple clients retrieving domain set data simultaneously (e.g. upon commit notify), could result in a server crash due to a race condition introduced by the branch-dependent domain set filtering added in CDM version 1.11
Newly added domain set elements during
cdm::CommitElementInCategory()are now restricted to the specified element in the specified set only
- 184.108.40.206 [11-07-2019]
Added client and service instance ids to improve service logging and matching of service and client log files
Improve dump file creation on-premise
- 220.127.116.11 [09-07-2019]
Added support for new
Added support creating of dump files (on-premise) or core dumps (cloud platform)
- 18.104.22.168 [16-06-2019]
When domain set membership tables were stored in a category checked-out from a different branch than the categories containing identifier data dependent on these domain sets, checking out the data category containing such identifiers would result in empty data. Now, when checking out, identifier data will be filtered against the active set elements of domain sets with regard to the checked-out branch of the categories containing such domain sets.
When upgrading older CDM servers to more recent versions, the naming of truncated column names longer than the maximum column name length supported by the backing database could be changed depending on the deployment platform and compiler used to create the CDM server executables, leading to errors when checking out or committing data from such old databases. During the upgrade to version 22.214.171.124 or beyond, the existing truncated column names will now be stored in an additional column of the intrinsic CDM data definition table and used during data transfer. This will upgrade the CDM database version key. After the CDM database upgrade, the original CDM servers will still be able to use such upgraded CDM databases as before.
Negative integer labels could erroneously be translated to unmapped labels from other sets, leading to data being stored for incorrect tuples, and possibly to duplicate tuple error during commits.
Category ordering algorithm could lead to incorrect ordering in the presence of defined subsets that were artificially included in the identifier ordering to help the CDM dll to update such subsets when needed during checkouts.
This build will no longer support Win32 AIMMS versions
- 126.96.36.199 [20-05-2019]
Reading data for integer sets could cause a crash
- 188.8.131.52 [14-05-2019]
Added support for release notes
- 184.108.40.206 [11-05-2019]
Improve performance by not unnecessarily pulling category data upon new commit notifications when categories were already at the latest revision. Note that the commit notification procedures have gotten a new optional 4th argument, which is required for this performance improvement to work. If you have implemented a custom commit notification procedure, then you should add the 4th argument and re-visit
cdm::DefaultCommitInfoNotification()to investigate what further changes to your custom commit notification procedure are required.
- 220.127.116.11 [25-04-2019]
On-demand CDM service in AIMMS cloud could hang on exit, leading to new clients not being serviced properly
- 18.104.22.168 [24-04-2019]
Automatic conversion of string to int did not work on all databases in
cdm::EmptyElementInCategory()could assign empty value to non-existing tuple.
Records of snapshot revision in
cdm::RetireBranchData()had ids potentially greater than ids of later revisions on same branch, leading to erroneous checkout results.
- 22.214.171.124 [23-04-2019]
MSOBDCSQL13 driver for SQLServer did not accept automatic conversion from integer to string in
- 126.96.36.199 [11-04-2019]
Added DLL that was preventing CDM from being run from Windows PRO client
- 188.8.131.52 [09-04-2019]
cdm::Branches()set elements were determined with respect to incorrect set in
- 184.108.40.206 [05-04-2019]
Added capability to retire intermediate commits by a single snapshot, via
Modified code to use non-persistent intermediate tables for storing current set content when checking out data to speed up checkout.
cdm::RevisionIdentifierCard()identifier, holding per-revision cardinality of changes for each individual identifier.
When checking out data, cleanup
cdmrt::criidentifiers in addition to emptying, in case domain sets have been cleared which might leave inactive data behind.
- 220.127.116.11 [27-03-2019]
Added capability to compare branches via
cdmrt::bci()shadow identifiers, and
- 18.104.22.168 [04-03-2019]
cdm::CloneAndRollbackElementInCategory()for integer sets where integer master set (i.e. not root set) is not in the category to which the function is applied.
Inactive data due to inactive domain set elements could lead to delta out of identifiers with such inactive data not to be stored, and consequently the commit to be only partial.
Element parameter with default that was not (by coincidence) an integer, lead to database query errors, because of not being translated to label number in all cases.
Modified code to circumvent MSOBDCSQL13 driver problem.
Modified code to skip unresolvable tuples when handling incoming changes for multi-dimensional identifiers, and log the corresponding offending label names, instead of skipping the entire assignment to the model identifiers
- 22.214.171.124 [09-10-2018]
- 126.96.36.199 [12-09-2018]
Added support for VS2017 builds of AIMMS.
- 188.8.131.52 [06-09-2018]
Fixed foreign key constraint problem when deleting branches
Modified code to catch connection lost exceptions and report properly to the model
- 184.108.40.206 [26-07-2018]
- 220.127.116.11 [09-05-2018]
Modified code to support relative tolerance when comparing values
- 18.104.22.168 [15-04-2018]
Initial public release of the CDM library