fix(lake-formation-tag-sync): option to keep connection key in schema name

[christopher-tin-atlan]

What

Lake Formation Tag Sync builds each schema qualifiedName from the Lake Formation DatabaseName with its first _-separated segment (the connection_map.json key) stripped off: prod_sales_analytics → key prod, schema sales_analytics.

That's correct when the crawled schema name excludes the prefix. When the crawled schema keeps it (prod_sales_analytics), every schema, table and column lookup misses and is skipped as not found in update-only mode. No connection_map.json value can work around it: the code always puts / between the mapped connection and the stripped remainder, and an empty-string key can never match.

Separately, a DatabaseName with no _ throws IndexOutOfBoundsException when the split result is destructured.

How

• New opt-in input keep_database_prefix ("Keep connection key in schema name"), default false. When true, the first segment still resolves the connection, but the full DatabaseName is used as the schema name. remove_schema strips whichever schema name is in use.
• A DatabaseName with no _ is logged and skipped instead of throwing.
• Help text: fixed the "Drom" typo on remove_schema and gave both schema-name options a concrete example; the keep_database_prefix help states the default (the connection_map.json key is stripped from the schema name). Labels and keys are unchanged, but the remove_schema help text changes for every tenant.
• LakeFormationTagSyncCfg.kt regenerated from package.pkl via genCustomPkg.

Default behaviour is unchanged, so existing tenant configurations that rely on the stripping keep working.

Testing

New CSVProducerSchemaNameTest (no tenant needed), 4/4 passing locally on JDK 17:

• default still strips the key from the schema
• flag on uses the full DatabaseName for schema, table and column qualifiedNames
• flag on with remove_schema strips the full name from the table name
• no-underscore DatabaseName is skipped, not thrown

Existing tenant-backed tests (CSVProducerTest, LakeTagSynchronizerTest, EnumCreatorTest) were not run locally.

Follow-up

After release, bump the csa-lake-formation-tag-sync image and add the new field in marketplace-packages/packages/csa/lake-formation-tag-sync.

Refs CSA-628

🤖 Generated with Claude Code

[linear]

CSA-628

[christopher-tin-atlan]

End-to-end verification

Tested on an internal Atlan tenant with a branch image built through the manual merge.yml run (7.4.2-SNAPSHOT-christophertin-csa-628). The feed had two association files against one connection:

• key_schema_a, where the crawled schema name keeps the key prefix;
• otherkey_key_schema_a, the old shape, where the key is stripped.

	Run 1: option absent from the saved workflow	Run 2: keep_database_prefix: true
NESTED_CONFIG	"keep_database_prefix": false, from the template default	"keep_database_prefix": true
Prefix kept in the crawled schema name	Looked up …/schema_a, 33 skipped (reproduces the bug)	Looked up …/key_schema_a, 1 schema + 1 table + 31 columns updated, 0 skipped
Old shape	33 matched, unchanged	Skipped, as expected with the option on

After run 2 I confirmed in the catalog that all 33 assets carry the value written by the run.

Backward compatibility: a workflow saved before this input existed ran unchanged. Argo applied the template default, and LakeFormationTagSyncCfg defaults to false when the key is missing.

Unrelated issue found while testing: CustomMetadataCache fails to load when any custom-metadata attribute on the tenant has options.primitiveType outside AtlanCustomAttributePrimitiveType (e.g. "text"):

InvalidFormatException: Cannot deserialize value of type AtlanCustomAttributePrimitiveType from String "text"

That stops every package that loads the cache. Not addressed here; I'll raise it separately.

[christopher-tin-atlan]

Run 1 (backwards compatibility): https://fs3.atlan.com/workflows/profile/csa-lake-formation-tag-sync-1768497496/runs?name=csa-lake-formation-tag-sync-1768497496-pdllf

Run 2 (showing skip due to new logic): https://fs3.atlan.com/workflows/profile/csa-lake-formation-tag-sync-1768497496/runs?name=csa-lake-formation-tag-sync-1768497496-68ghc

[jtgatlan]

Nice work on this, Tin. The fix is clean and keeping it opt-in protects existing configs.

One thing before merge: the green pr-test check doesn't actually cover the new CSVProducerSchemaNameTest. Package tests only run when -PpackageTests is set (see buildSrc/src/main/kotlin/com.atlan.kotlin-custom-package.gradle.kts), so CI skipped it. Could you paste the output of your local run here, or run:

./gradlew :samples:packages:lake-formation-tag-sync:test --tests "com.atlan.pkg.lftag.CSVProducerSchemaNameTest" -PpackageTests

That way the PR has a record that the tests passed. Thanks!

[christopher-tin-atlan]

Thanks John, good catch on -PpackageTests. Here's a fresh local run of that command against the PR head (019a812f8d), with --rerun-tasks so nothing came from cache. JDK 17 (OpenJDK 17.0.20.1):

$ ./gradlew :samples:packages:lake-formation-tag-sync:test --tests "com.atlan.pkg.lftag.CSVProducerSchemaNameTest" -PpackageTests --rerun-tasks

com.atlan.pkg.lftag.CSVProducerSchemaNameTest databaseNameWithoutUnderscoreIsSkippedNotThrown PASSED
com.atlan.pkg.lftag.CSVProducerSchemaNameTest defaultStripsConnectionKeyFromSchema PASSED
com.atlan.pkg.lftag.CSVProducerSchemaNameTest keepDatabasePrefixUsesFullDatabaseNameAsSchema PASSED
com.atlan.pkg.lftag.CSVProducerSchemaNameTest keepDatabasePrefixWithRemoveSchemaStripsFullDatabaseName PASSED
SUCCESS: Executed 4 tests in 911ms
BUILD SUCCESSFUL in 2m 3s

These tests don't need a tenant: they call CSVProducer.transform() directly and check the qualifiedNames it writes. The existing tenant-backed tests (CSVProducerTest, LakeTagSynchronizerTest, EnumCreatorTest) weren't run locally; the end-to-end runs on the test tenant above cover that path.

[jtgatlan]

Thanks Tin, the local run plus the end-to-end runs on the test tenant (including the backward-compat check) give us everything we need. Approving.

Two small things for after merge: the marketplace-packages bump so the new option shows up for users, and a Linear ticket for the CustomMetadataCache primitiveType issue you found, since that one can stop every package on an affected tenant. Great work on this.