Error index

Applies to VAST Cluster 5.5.

The errors you can get when you create, load, change and search vector-indexed tables, with their causes and fixes. Search this page for a distinctive part of your message, such as must be a vector column or Cannot cast.

How the messages were produced: every message on this page was captured on a VAST Cluster 5.5 system (5.5.0-SP3 or 5.5.0-SP6) through the VAST Query Engine ADBC driver (adbc-driver-vastdb 0.0.32) or the Python SDK 2.1.0, or, for the VMS REST API rows, with curl. <placeholders> mark values that vary, such as column names and request IDs.

Reading a Query Engine error

ADBC errors from the Query Engine come in two shapes. Look for the text after failed with: or description::

... failed with: Binder Error: Vector index column \"name\" must be a vector column, but has type: VARCHAR", request_id: "<id>" }
... description: [req=0x<id>]: internal error: Invalid Argument)

(Abridged at ....)

The first shape names the problem. The second is a rejection by the database with no detail. For Invalid Argument, use the "when" column in the tables below to find which change was refused. Python SDK errors are vastdb.errors.HttpError subclasses with code and message fields (reading an SDK error).

Connecting

Message	Cause	Fix
`Request GetVersionRequest failed ... error_code: Other("SignatureDoesNotMatch"), message: "The request signature we calculated does not match the signature you provided. Check your AWS secret access key and signing method."`	The secret key is wrong for this access key.	Check `vast.db.secret_key`. See setup.
`vastdb.transaction.NoAdbcConnectionError: Adbc Driver may not have been supplied`	The SDK's transaction opened without the ADBC driver, and you then called `vector_search()`.	`pip install 'vastdb[adbc]'`. See VAST DataBase errors.
`Network error: http://<addr>/?transaction=: Connection Failed: Connect error: Connection refused (os error 111)`	The endpoint URL has the wrong port, or none: the Query Engine answers on the cluster's S3 port, such as `:9090`.	Add the port to `vast.db.endpoint` (setup).
`Query stopped prematurely: Query canceled: Server error (status: 17, description: We encountered an internal error. Please try again.)` on every query, while `CREATE SCHEMA` and `CREATE TABLE` succeed	The cluster has no virtual IP pool with the Query Engine role. Until one exists, DDL works but queries fail on every virtual IP.	Ask your platform team for a Query Engine pool, and connect to it (setup).
`Conflict: {'code': 'InvalidBucketState', 'message': '[req=0x<id>]: This @s3_bucket_name=<bucket> does not support database api - failing request' ... 'status': 409`	Python SDK: the bucket is on an S3-only view, not a Database-enabled view. Through ADBC the same mistake gives only `Catalog Error: Table with name <table> does not exist!`	Use a bucket on a Database-enabled view (setup).
`Binder Error: [req=0x<id>]: permission checking for table: <bucket>/<schema>/<table> failed - AccessDenied.`	ADBC, or the SDK's `vector_search()`: the user's policy doesn't allow `s3:QueryEngine*`. The SDK's own `insert()` and `select()` still work without it.	Ask for `s3:QueryEngine*` (setup).
`Query Status Keepalive Failure: Get query status request failed. (Request GetQueryStatusRequest(...) failed Server responded with error: ServerErrorResponse { error_code: Other("AccessDenied"), message: "[req=0x<id>]: request was denied according to policies res=2" ...` on every query, even `SELECT 1`, while the SDK's `insert()` and `select()` work	No identity policy applies to your user in your tenant. Usually the policy was attached in the default tenant: for a user of the cluster's default local provider, that's where it goes unless the tenant is named. The SDK's own calls still work for a user that owns the bucket.	Ask your platform team to attach the policy to your user in your tenant: `user modify --tenant-id`, or `PATCH /api/users/<user id>/tenant_data/` with `tenant_id` (setup).
`AccessDenied ... [req=0x<id>]: login_name=<user> was not found`	The `vast.db.end_user` connection option names a user the cluster doesn't know	Pass the end user's VAST user name (security).

Creating a table

Message	When	Fix
`Binder Error: Vector index column "<col>" must be a vector column, but has type: VARCHAR` (or `INTEGER`)	`VECTOR CLUSTER BY` names a column that isn't a vector	Declare the column as `VECTOR(FLOAT, <d>)` (or `VECTOR(FLOAT16, <d>)`) and cluster by it.
`Binder Error: Vector index column "<col>" must be a vector column, but has type: FLOAT[]`	The column is a list (`FLOAT[]`), not a vector	Use `VECTOR(FLOAT, <d>)`.
`Binder Error: ARRAY type is not supported in table definitions. Column "<col>" has type: FLOAT[4]` (or `INTEGER[4]`)	A fixed-size array type in `CREATE TABLE`	Use `VECTOR(FLOAT, <d>)` for the column. Fixed-size arrays such as `FLOAT[4]` are for casts in queries, not for columns.
`Parser Error: Duplicate attribute: 'metric'`	`metric` given twice in `VECTOR CLUSTER BY ... (...)`	Give each option once.
`Parser Error: syntax error at or near "vector"`	A column named `vector`, unquoted. The word is reserved in the Query Engine.	Rename the column, for example `embedding`. A quoted `"vector"` is accepted, but don't use the name.
`Parser Error: syntax error at or near "SECONDARY"` or `... near "INCLUDE"`	`SECONDARY INDEX` or `INCLUDE (...)` clauses, which 5.5.0 doesn't support	Remove the clause. Filter with `WHERE` instead (filters).
`internal error: Invalid Argument`	Through SQL: an unknown `metric`, an unknown quantization value (including `rabitq_*`), or the pair `cluster_quantization = 'scalar_8bit'`, `search_quantization = 'none'`	Use `metric` `l2sq`, `cosine` or `ip`, and quantization `none` or `scalar_8bit` with a legal pair (quantization). The SDK messages below say which.
`BadRequest ... 'code': 'InvalidArgument', 'message': "[req=0x<id>]: search_quantization='none' requires cluster_quantization='none'"`	Python SDK: `VectorIndexSpec(..., cluster_quantization="scalar_8bit", search_quantization="none")`	Use `none`/`none`, `none`/`scalar_8bit` or `scalar_8bit`/`scalar_8bit`.
`BadRequest ... "[req=0x<id>]: invalid vector index cluster quantization '<value>', supported values: 'none', 'scalar_8bit'"` (or `search quantization`)	Python SDK: an unsupported quantization value, such as `rabitq_4bit`	Use `none` or `scalar_8bit`.
`BadRequest ... "[req=0x<id>]: invalid vector index distance metric <metric>, supported metrics: 'l2sq', 'ip', 'cosine'"`	Python SDK: an unsupported `distance_metric`	Use `l2sq`, `ip` or `cosine`. The SDK 2.1.0 docstring lists only `l2sq` and `ip`, but the cluster accepts `cosine` too.
`BadRequest ... "[req=0x<id>]: invalid vector indexed column name <col>"`	Python SDK: `VectorIndexSpec(column=...)` names a column that isn't in the table's schema	Pass the name of the vector column.
`NotImplemented ... vector indexing cannot be combined with elysium sorting` (or `... with partitioning`)	Python SDK: `create_table()` with a vector index and a sort key or partitions. "Elysium" in the message means sorted tables.	A vector-indexed table can't be sorted or partitioned (combining table structures).
`failed to enable vector index eres=106`	Python SDK: `create_table()` with a vector index and user-defined row IDs	The two can't be combined (combining table structures).
`BadRequest: {'code': 'InvalidArgument', 'message': '[req=0x<id>]: failed to enable vector index eres=14' ... 'status': 400`	Python SDK: `create_table()` with a vector index and a `vastdb_rowid` column	Don't add `vastdb_rowid` to an indexed table. The failed call still creates the table, without an index: drop it before you try again.
`{"detail":"2 validation errors for TableInputSchema\nvector_indexes -> 0 -> cluster_quantization\n field required (type=value_error.missing)\nvector_indexes -> 0 -> search_quantization\n field required (type=value_error.missing)"}`, HTTP 400	VMS REST API `POST /tables/`: a `vector_indexes` entry without the quantization fields (here, both are missing). The REST API has no default quantization.	Send both `cluster_quantization` and `search_quantization` (CI/CD and automation).
`{"detail":"1 validation error for TableInputSchema\n__root__\n search_quantization=NONE requires cluster_quantization=NONE (type=value_error)"}`, HTTP 400	VMS REST API: `SCALAR_8BIT` for clustering with `NONE` for search	Use `NONE`/`NONE`, `NONE`/`SCALAR_8BIT` or `SCALAR_8BIT`/`SCALAR_8BIT` (CI/CD and automation).
`[req=0x<id>]: vector dimension 2148 exceeds max supported dimension 2147 for clustering quantization type SCALAR_8BIT due to memory limitations` (or `... 536 for clustering quantization type NONE ...`)	The vector column has more dimensions than the index supports for its cluster quantization: 2,147 for `scalar_8bit`, 536 for `none`	Use `cluster_quantization = 'scalar_8bit'`, or a model with fewer dimensions (limits). A failed Python SDK `create_table()` can leave a table behind that answers `DROP` with `404 NoSuchKey`; list the schema's tables before you retry.

Inserting rows

Message	When	Fix
`Conversion Error: Cannot cast array of size 7 to array of size 8`	Inserting a value cast to `FLOAT[7]` into a `VECTOR(FLOAT, 8)` column	Every vector must have exactly the column's dimension. Check the embedding model's output size.
`Conversion Error: Cannot cast list with length 3 to vector with length 4`	Inserting a list literal, such as `[1.0, 2.0, 3.0]`, of the wrong length	Same as above.
`Conversion Error: Cannot cast list with NULL values to vector (non-nullable nested type)`	Inserting a list literal that contains a `NULL` element	Remove or replace the missing values before you insert.
`internal error: Invalid Argument`	Inserting a `NULL` vector, a vector containing `NaN` or infinity, or a `FLOAT16` vector with a value outside the `FLOAT16` range (it would become infinity)	Give every row a vector of finite values, and keep `FLOAT16` values in range, or use `VECTOR(FLOAT, <d>)`.
`BadRequest: {'code': 'InvalidArgument', 'message': '[req=0x<id>]: vector index column contains non-finite values (has_nan=1 has_inf=0)' ... 'status': 400`	Python SDK `insert()`: a vector containing `NaN` or infinity	Clean the vectors before you insert.
`Server error (status: 17, description: [req=0x<id>]: [req=0x<id>]: internal error: We encountered an internal error. Please try again.)`	`INSERT` into a table whose indexed column is `VECTOR(INTEGER, <d>)`. The table is created, but rows can't be inserted through SQL.	Use `VECTOR(FLOAT, <d>)` or `VECTOR(FLOAT16, <d>)`.
`InternalError ... Logical plan is too big: <size> > 4194304`	A multi-row `INSERT ... VALUES` with vector literals: the query plan outgrows its size limit before the statement reaches the 5 MB cap	Insert fewer rows per statement, or load with bulk ingest (ingest).
`RequestDataTooLarge ... Content_length=<n> is larger than max available buffer cmd=START_QUERY`	The SQL statement is larger than the 5 MB cap (limits)	Same as above.

A NULL element inside a value that's explicitly cast, such as [1, 2, 3, NULL]::FLOAT[4], is accepted and stored as 0.0. Check your vectors for missing values before you load them.

Changing a table

Message	When	Fix
`internal error: Invalid Argument`	`UPDATE` of the vector column	Delete the row and insert it again with the new vector. Updating other columns works.
`internal error: Invalid Argument`	`ALTER TABLE ... ADD COLUMN` or `DROP COLUMN` on a vector-indexed table	Adding or dropping columns on an indexed table isn't supported in 5.5. Plan the columns before you create the table, or copy into a new table (move vectors into an indexed table).
`BadRequest ... InvalidArgument ... adding columns on indexed tables is forbidden` (or `dropping columns ...`)	Python SDK: `add_column()` or `drop_column()` on a vector-indexed table	Same as above (VAST DataBase errors).
`internal error: Tried to drop a column which cannot be dropped`	`ALTER TABLE ... RENAME COLUMN` on the indexed vector column	The indexed vector column can't be renamed.

Searching

Message	When	Fix
`AssertionError: Table is either not vector indexed. (maybe try reloading the TableMetadata)`	Python SDK: `vector_search()` on a table without a vector index, or with stale table metadata	Check that the table was created with a vector index. If it was, reopen the table in a new transaction.
`The operator 'SET' type=69 is currently not supported.` (code `NotImplemented`)	A SQL `SET` statement, for example to change a probe setting	Pass settings as ADBC connection options instead (configuration).
`The operator 'ORDER_BY' type=1 is currently not supported for this usage.` (code `NotImplemented`)	An `ORDER BY` a distance with no `LIMIT`	Add `LIMIT k`. A search should always have one (query design).
`Catalog Error: Scalar Function with name cosine_similarity does not exist!` (code `QueryCatalogException`)	SQL copied from a slide or webinar. The Query Engine has no `cosine_similarity` function.	Use `array_cosine_distance(...)` with `ORDER BY ... ASC LIMIT k` (query design). Don't rewrite it as `array_cosine_similarity(...)` with `ORDER BY ... DESC`: that runs, but as a full scan.
`Binder Error: Referenced column "<col>" not found in FROM clause!` for a column the table has	Column security denies you that column. Denying the vector column blocks every search on the table, including `vector_search()`; denying a payload column blocks any query that selects it.	Ask for access, or select only the columns you're allowed (security).
`Permission Error: [req=0x<id>]: query_engine is not supported for tables with row_filter=<mask>` (code `AccessDenied`)	A row filter applies to you on this table. The Query Engine doesn't run on a table with a row filter for that user, even if the filter is scoped to Trino.	See security.

A search that doesn't match the index's query shape usually doesn't fail: it runs as a full scan. See troubleshooting.