Previously we were using `NoopHostnameVerifier` which turns off hostname verification,
if the HostnameVerifier is not set it uses `DefaultHostnameVerifier` which performs
hostname verification

Co-authored-by: Jan Waś <[email protected]>

Additionally disable local scheduling of Mongo splits

Materialized view properties can be more expensive to retrieve
and are unused except for SHOW CREATE queries. Splitting them
from the main definition reduces file system calls for metadata
queries against the Iceberg connector.

Multiple modules depend on org.threeten.bp.Duration.
Move to the root pom.

Cleanup

This change will help in reducing memory
footprint in long-running partitioned
writes. Thus, we can scale more to increase
the write performance.

After the initial KMP mismatch, the next match was being skipped
at position i == longest match.

…tly"

This needs more though and a different approach that does not involve modifying the
information_schema tables.

This reverts commit 1c5e5eb.

CREATE TABLE with supported type is tested in type mapping tests already

The concrete type declared in the JsonTableFunctionHandle is not directly serializable.
The deserializer binding is only declared for the Type class.

PagePartitioners should either flatten their dictionary mode appenders
into direct mode, or force flush their current page to preserve the
dictionary encoding before being released to the pool for reuse since it
is not possible for the appender to observe the same dictionary input
when used from a different driver and the dictionary appenders do not
accurately report their size and therefore may have been preventing a
flush from occurring up until this point.

Also fixes an issue where dictionary block outputs were severely under
reporting their output size in bytes.

Spark sets the table property NESSIE_COMMIT_ID_PROPERTY in
NessieTableOperations#loadTableMetadata.
Then NessieIcebergClient.commitTable uses this property.

In Trino, this property is never set but used in
NessieIcebergClient.commitTable as it is a common code.
Hence, the commit id is old and doesn't allow new commits.

Use the common code (available From Iceberg 1.4.0)
NessieUtil.updateTableMetadataWithNessieSpecificProperties in Trino,
which handles setting the property like "nessie.commit.id".

Filter for processing the files which are smaller than
the threshold and consider a partition for optimization
if and only if it contains at least two such files.

Fixes negative cost estimated failure in
`TestQueryPlanDeterminism.testUnionAllAboveBroadcastJoin`.

In `LocalQueryRunner` we were assuming 0 "node count for stats", and
getting negative cost in the formula (somewhere deep in the stats
calculator):

    double cpuCost = buildSideSize * (estimatedSourceDistributedTaskCount - 1);

where `estimatedSourceDistributedTaskCount` is the node count.

Current formulas like

    double cpuCost = buildSideSize * (estimatedSourceDistributedTaskCount - 1);

would produce negative values when node count is 0.
In fact, 0 nodes cannot execute a query, so 0 is not a valid #nodes
estimate.

No work has negative cost. Previously such negative costs could go
unnoticed, or noticed only if explain plan is produced.

The build was observed failing with

    Warning: [325.033s][warning][gc,alloc] mvn-builder-trino-parquet: Retried waiting for GCLocker too often allocating 4194306 words

on CI.

Before the change, `FileHiveMetastore` would rely on
`fileSystem.listFiles` to find schemas or tables within a schema.
`listFiles` is recursive and, on a local file system, fails when
files/directories are being modified concurrently. This commit replaces
recursive `listFiles` listing with non-recursive `listDirectories`
leveraging the fact that listed entities are represented as directories.

The test sometimes failed due to a timeout on CI.

Results should be identical, but the new code is easier to extend for failover.

Convert product tests launcher's test to JUnit. This does not change
product tests themselves.

Remove TestNG from classpath of modules that do not use it anymore.

Add explicit mention what connector support storage so readers dont have
to go looking through all connectors to find out.

The exception is due to the value user entered hence should be reported as a TrinoException rather than a GENERIC_INTERNAL_ERROR

When defining test profiles, we don't use properties.

`GcsTransactionLogSynchronizer` requires that the new transaction log file
is created exclusively and atomically. Otherwise concurrent readers of a
table may see a file partially written (or not written yet, just empty)
and fail. This commit:

- fixes the GCS native filesystem implementation so that it's atomic
- changes the method signature to indicate atomic creation and
  remove default not atomic implementation.
- makes it clear in-memory buffering occurs (previously it was
  implicitly done in `HdfsOutputFile` which could be considered
  surprising)
- in `AbstractTestTrinoFileSystem` decouples "is create() exclusive" and
  "supports createExclusive" behaviors. For example local filesystem has
  the former, GCS filesystem has both and S3 filesystem has none.

Engine-provided functions do not need
`ClassLoaderSafeConnectorTableFunction` wrapper.

Engine-provided functions do not need
`ClassLoaderSafeConnectorTableFunction` wrapper.

When `ConstantExpression` was created with a Trino type and a value
that didn't match this type's java type (for example Trino INTEGER with
value being Integer, instead of Long), the `ConstantExpression` would
fail to serialize to JSON. Such failure happens during sending task
status updates to workers and is currently logged and ignored, leading
to query hang. The problem could be triggered with SQL routines, which
utilize `RowExpression` (including `ConstantExpression`) serialization.

Thus, this fixes execution of SQL routines involving Row field
dereference.

The test must run single-threaded and was run like that until
9a7cf10.

Query start may be time consuming operation and should be traced.
Especially in case of `DataDefinitionExecution`, the start may cover all
of the query execution.

Tests extending `AbstractTestQueryFramework` inherit `PER_CLASS`
lifecycle and `CONCURRENT` behavior and that should be expected given
how `AbstractTestQueryFramework` works. No need to define that in
subclasses, and indeed most subclasses do not define these behaviors
again.

We observed duplicate attempts to drop spoolingOutputStats in some logs.  These warnings will help diagnose what's going on.

Some tests being fixed simply fail because they do not support
concurrent test method execution.
Others worked, but could allocate lots of resources, potentially leading
to OOM failures on CI.

The tests where found by looking at
9a7cf10 commit:

    git show 9a7cf10 | grep -A 3 '@test(singleThreaded = true)'

`TestQueryManager` defined a default `queryRunner` used by one test
method, but other test methods created additional runners. This commit
moves the default one into the only test method that used it. In the
result, there is at most 1 query runner at the same time created during
test execution.

Sometimes, `ConnectorSplitManager.getSplits` can take long to construct
`ConnectorSplitSource`. For example, in Delta, there is IO work being
done before `ConnectorSplitSource` is returned. This work would better
be delayed until `ConnectorSplitSource.getNextBatch` is invoked, but
currently this is not the case. Let's add tracing so that time spent in
`ConnectorSplitManager.getSplits` is attributable.

The parameter refers to "a stage span", not to the top level "query
span".

Failure injection is a for-tests functionality. Hide its config, they
are not meant to be used.

The change is to make the schema sync with hudi_cow_pt_tbl.

Catch exception in sendUpdate. Currently in case we get unexpected
exception from that method query would fail as thread responsible for
updating task state will terminate.