Add support for case sensitive identifiers #17

martint · 2019-01-21T03:40:24Z

<delimited identifier> ::=
  <double quote> <delimited identifier body> <double quote>

<delimited identifier body> ::=  <delimited identifier part>...
<delimited identifier part> ::=
    <nondoublequote character>
  | <doublequote symbol>

<Unicode delimited identifier> ::=
  U <ampersand> <double quote> <Unicode delimiter body> <double quote>
      <Unicode escape specifier>
<Unicode escape specifier> ::=
  [ UESCAPE <quote> <Unicode escape character> <quote> ]
<Unicode delimiter body> ::=
  <Unicode identifier part>...
<Unicode identifier part> ::=
    <delimited identifier part>
  | <Unicode escape value>

24) For every <identifier body> IB there is exactly one corresponding case-normal form CNF. CNF is an <identifier body> derived from IB as follows:
Let n be the number of characters in IB. For i ranging from 1 (one) to n, the i-th character Mi of IB is transliterated into the corresponding character 
or characters of CNF as follows:
Case:
   a) If Mi is a lower case character or a title case character for which an equivalent upper case sequence U is de ned by Unicode, then let j be th
       e number of characters in U; the next j characters of CNF are U.
   b) Otherwise, the next character of CNF is Mi.
25) The case-normal form of the <identifier body> of a <regular identifier> is used for purposes such as and including determination of identifier 
      equivalence, representation in the Definition and Information Schemas, and representation in diagnostics areas.

...

27) Two <regular identifier>s are equivalent if the case-normal forms of their <identifier body>s, considered as the repetition of a <character string literal> 
that specifies a <character set specification> of SQL_IDENTIFIER and an implementation-defined collation IDC that is sensitive to case, compare equally 
according to the comparison rules in Subclause 8.2, “<comparison predicate>”.

28) A <regular identifier> and a <delimited identifier> are equivalent if the case-normal form of the <identifier body> of the <regular identifier> and the 
<delimited identifier body> of the <delimited identifier> (with all occurrences of <quote> replaced by <quote symbol> and all occurrences of 
<doublequote symbol> replaced by <double quote>), considered as the repetition of a <character string literal> that specifies a <character set specification>
 of SQL_IDENTIFIER and IDC, compare equally according to the comparison rules in Subclause 8.2, “<comparison predicate>”.


29) Two<delimited identifier>s are equivalent if their <delimited identifierbody>s,considered as the repetition of a <character string literal> that specifies
 a <character set specification> of SQL_IDENTIFIER and an implementation-defined collation that is sensitive to case, compare equally according to the
 comparison rules in Subclause 8.2, “<comparison predicate>”.

30) Two <Unicode delimited identifier>s are equivalent if their <Unicode delimiter body>s, considered as the repetition of a <character string literal> that
 specifies a <character set specification> of SQL_IDENTIFIER and an implementation-defined collation that is sensitive to case, compare equally according
 to the comparison rules in Subclause 8.2, “<comparison predicate>”.

31) A <Unicode delimited identifier> and a <delimited identifier> are equivalent if their <Unicode delimiter body> and <delimited identifier body>, 
respectively, each considered as the repetition of a <character string literal> that specifies a <character set specification> of SQL_IDENTIFIER and 
an implementation-defined collation that is sensitive to case, compare equally according to the comparison rules in Subclause 8.2, “<comparison predicate>”.

32) A <regular identifier> and a <Unicode delimited identifier> are equivalent if the case-normal form of the <identifier body> of the <regular identifier> 
and the <Unicode delimiter body> of the <Unicode delimited identifier> considered as the repetition of a <character string literal>, each specifying a
 <character set specification> of SQL_IDENTIFIER and an implementation-defined collation that is sensitive to case, compare equally according to the 
comparison rules in Subclause 8.2, “<comparison predicate>”.

The approach and design is being captured here: https://github.com/prestosql/presto/wiki/Delimited-Identifiers

The text was updated successfully, but these errors were encountered:

martint · 2019-03-04T20:05:34Z

We need to decide on what (if anything) we want to do for backwards compatibility around this change. There are some points for consideration:

The engine needs to communicate to connectors whether names are delimited or non-delimited. This requires changing how table names, schema names, user names, roles, etc., are represented from a String to a wrapper class (Name?) that contains the value and whether it's delimited or not.
There are some methods in the connector SPIs that return List<String> (e.g., List<String> listSchemaNames(ConnectorSession session))
In the long term, we want the "good" method names to be used for the implementation that supports delimited/non-delimited names. We don't want to get stuck with all the APIs having some long-winded or non-obvious method names because the current APIs already took them for what will eventually be the legacy implementation. For example, SchemaTable has a String getCatalog() method. The new method would need to be named differently, such as getOriginalCatalog() if we wanted to preserve backward compatibility.

So, some options to approach this change:

Add new parallel "legacy" APIs by copying the existing APIs and mark the current ones as deprecated. Connectors that don't want to support the new API can just do a simple rename to use the newly introduced "legacy" versions. After a some grace period, replace the current APIs with the variant that understands delimited names. Eventually, remove all the legacy APIs.
Add new APIs and mark the current ones as deprecated. Connectors can migrate to new APIs at will. Eventually, the current APIs get removed.
Forgo backward compatibility and update all APIs to support delimited names.

Option 1) satisfies all the considerations listed above, but requires connectors to make a change at some point, even to keep the current behavior. Updating the connector implementations to understand delimited names is a mostly mechanical affair. So, updating connectors to work against the newly introduced legacy APIs is a waste of effort.

Option 2) leaves us with the poor choice of names for the long term.

Option 3), well, doesn't give us backward compatibility. On the other hand, since the change is mostly mechanical, especially if the connector wants to keep existing semantics, it seems like this wouldn't be too much of a burden.

The more I think about it, the more I lean towards option 3. Thoughts? Any other alternatives I haven't listed?

findepi · 2019-03-05T09:31:14Z

Option 2) leaves us with the poor choice of names for the long term.

Note that you can extend option 2 -- after the transition period, when old APIs are removed, you can rename new APIs to have better method names.

Updating the connector implementations to understand delimited names is a mostly mechanical affair.

@martint i think JDBC connectors require some thought here. Consider SHOW TABLES FROM schema example.
If schema is delimited, this is actually easy, java.sql.DatabaseMetaData#getTables should do the trick.
If schema is non-delimited, you need to find the actual schema name (case-insensitively) before calling java.sql.DatabaseMetaData#getTables. (We actually have something like that implemented, and can share this)

For this reason, I think we need some kind of transition period. Although such a transition might be implemented in each connector individually (under option 3), i mean leaning towards option 1.

martint · 2019-03-05T16:25:29Z

One downside of one is that during the transition period we'd have an API that's marked as deprecated and one that's considered legacy. Which one should new connector writers use? Clearly, not the deprecated one. But using the legacy one feels silly ("why can't I use a non-legacy one?") and forces them to adjust their usages later.

On the other hand, as you pointed out, with option 3 every connector has the choice of migrating to support delimited identifiers or preserve the current semantics (treat everything as non-delimited as today) and deal with it later.

findepi · 2019-04-23T08:16:30Z

@Praveen2112 @martint @kokosing

i was thinking about this today and came to conclusion that there are these concepts:

object name (where object is a catalog, schema, table, view, role, function, user, etc.)
identifier name in a query (can be quoted or not)

We need to distinguish them, because they are different.

table name abc is just the three lower-case letters; a table name cannot decide whether it should be matched case-insensitively or not
identifier name in a query can be delimited (quoted) or not; query's "abc" matches table name abc, and query's abc matches table names abc, aBC, etc.

So far #354 apparently uses one class (Name) for both these concepts (symmetrically) and uses Name#equals to match them.
I would rather see these concepts distinguished in the code.

Proposed representation

use String (or simple String wrapper class Name(String name)) to represent object names
use NameSelector(String name, boolean delimited) to represent identifier name in a query
have NameSelector#matches(String objectName) (or NameSelector#matches(Name objectName)) to match identifier name in a query against object name

Examples:

ConnectorMetadata#schemaExists will take NameSelector
ConnectorMetadata#listSchemaNames will return List<String> (or List<Name>)
ConnectorMetadata#listTables will take String (or Name) -- i.e. resolved, actual schema name

martint · 2019-04-23T18:08:02Z

The spec talks about object names being identifiers. This is pervasive across the spec document:

Let T be the table defined by the <table definition> TD. Let TN be the <table name> simply contained in TD.

A table descriptor TDS is created that describes T. TDS includes:
a) The table name TN.

A base table descriptor describes a base table. In addition to the components of every table descriptor, a base table descriptor includes:
- The name of the base table.

The grantee is <authorization identifier> A.

The row type RT of the table T defined by the <table definition> is the set of pairs (<field name>, <data type>) where <field name> is the name of a column C of T and <data type> is the declared type of C.

However, it then describes the domain of the columns that represent object names in the INFORMATION_SCHEMA and DEFINITION_SCHEMA tables using:

SQL_IDENTIFIER domain

Define a domain that contains all valid <identifier body>s and <delimited identifier body>s.

CREATE DOMAIN SQL_IDENTIFIER AS CHARACTER VARYING (L) CHARACTER SET SQL_IDENTIFIER

This domain specifies all variable-length character values that conform to the rules for formation and representation of an SQL <identifier body> or an SQL <delimited identifier body>

And:

The representation of an <identifier> in the base tables and views of the Information Schema is by a character string corresponding to its <identifier body> (in the case of a <regular identifier>) or its <delimited identifier body> (in the case of a ). Within this character string, any lower-case letter appearing in a <regular identifier> is replaced by the equivalent upper-case letter, and any <doublequote symbol> appearing in a <delimited identifier body> is replaced by a <double quote>.

As an example, this is how the SQL_IDENTIFIER domain is used in the definition of the CATALOG_NAME base table in DEFINITION_SCHEMA:

CREATE TABLE CATALOG_NAME (
    CATALOG_NAME INFORMATION_SCHEMA.SQL_IDENTIFIER,
    CONSTRAINT CATALOG_NAME_PRIMARY_KEY
    PRIMARY KEY ( CATALOG_NAME )
)

This means that when presented in the contents of INFORMATION_SCHEMA and DEFINITION_SCHEMA tables, identifiers lose any indicators of whether they were delimited or not when the objects were created. This is ok, though, since the upper-case normalization above guarantees that if you used such identifiers in a query, they'd match the corresponding object even if they are surrounded with quotes.

It also hints to the fact that only the body of the identifiers is stored, but it's not explicit about it. The following statement does nothing to clarify this:

Where an <actual identifier> has multiple forms that are equal according to the rules of Subclause 8.2, “<comparison predicate>”, in ISO/IEC 9075-2, the form stored is that encountered at definition time.

Unfortunately, it doesn't clarify whether the "form stored" includes the quotes, if present.

The book "SQL: 1999: Understanding Relational Language Components" by Jim Melton, one of the editors of the SQL spec says:

In effect, SQL:1999 changes all lowercase letters in regular identifiers to their uppercase-equivalent letters. This is especially important for identifiers that are stored as data values in the views of the Information Schema.
[...]
When a delimited identifier is stored into the views of the Information Schema, the double quotes are not stored, but all the other characters are, just as they appeared in the delimited identifier. Therefore, the regular identifier TITLES and the delimited identifier "TITLES" are stored identically and are therefore completely equivalent.

PostgreSQL behaves exactly like that, except that it normalizes to lower-case before storing, opposite of what the spec describes.

So, I think you're partly right in that, in theory, all we need when connectors present a name to the engine is a string. However there are some things to consider:

NameSelector may not be an appropriate concept for all usages. When creating a table, how do we convey the table and column names to the connector? We could use a string and normalize the name according to the rules in the SQL spec before providing it to connectors, but that would break connectors that don't play by the SQL rules. We'd either have to 1) add "delimited" to name, which brings us back to square one 2) add yet another entity to represent a "sql identifier".
When connectors return a name to the engine, do they first need to normalize it according to SQL rules? This could be impossible for connectors like PostgreSQL, which normalizes the names in the opposite case.

findepi · 2019-04-23T21:08:21Z

@martint thanks for looking into this.

When connectors return a name to the engine, do they first need to normalize it according to SQL rules?

For existing objects, names should be returned as-is.
If the remote storage is case insensitive (like Hive is, right?), we could add some normalization here.
But for JDBC connectors we generally shouldn't.
Table names are case-sensitive in eg Postgres, MySQL, SQL Server or Oracle. It is only "less convenient" (requires "-delimiting) to create tables with case different than the default.

NameSelector may not be an appropriate concept for all usages. When creating a table, how do we convey the table and column names to the connector?

Good point. When creating a table, we want to pass "identifier name from a query" to the connector.
We can pass the value normalized to lower- (or upper-) -case unless it was "-delimited.
(Even if we normalize to upper when talking to Postgres connector, it will still normalize to lower because the name is not delimited.)
Plus, we need to retain information whether it was delimited.

My envisioned NameSelector(String name, boolean delimited) fits here perfectly... except for its name.

So:

let's call this concept SqlIdentifier(String name, boolean delimited)
let's use it when looking for a table, or creating new table
let's use case-sensitive String (or Name, or ObjectName) for names of existing objects
let's use it when listing objects (eg listing schemas, listing tables)

Praveen2112 · 2019-04-24T14:46:18Z

@findepi Thanks for your insights.

let's call this concept SqlIdentifier(String name, boolean delimited)

But currently Name object also captures the same right i.e it maintains the rawName and the info whether it is delimited or not

kokosing · 2020-09-02T07:50:15Z

Just a reminder, that while fixing this issue we should also make sure that values returned from system catalog and information_schema have proper data and that predicate pushdown works for them. For example see: #4965

# This is the 1st commit message: Add Snowflake JDBC Connector # This is the commit message #2: Update trino snapshot version to 372 # This is the commit message trinodb#3: Update format of the doc of snowflake # This is the commit message trinodb#4: Update trino jdbc library import # This is the commit message trinodb#5: Fix date formatter from yyyy to uuuu # This is the commit message trinodb#6: Fix date test case # This is the commit message trinodb#7: Remove defunct property allow-drop-table # This is the commit message trinodb#8: Update trino version to 374 # This is the commit message trinodb#9: Update snowflake config to adapt 374 # This is the commit message trinodb#10: Update the range of the test of Date type # This is the commit message trinodb#11: Update to version 375 # This is the commit message trinodb#12: Fix snowflake after updating to 375 # This is the commit message trinodb#13: Update to 381 # This is the commit message trinodb#14: Fix mvn pom import # This is the commit message trinodb#15: Format snowflake.rst # This is the commit message trinodb#16: Reorderd Data tests in type mapping # This is the commit message trinodb#17: Update function code # This is the commit message trinodb#18: Add product test # This is the commit message trinodb#19: Rename product test tablename # This is the commit message trinodb#20: Add Env, Suite and properties of Snowflake for production test # This is the commit message trinodb#21: Add trinoCreateAndInsert() # This is the commit message trinodb#22: Refactor snowflake from single node to multi node # This is the commit message trinodb#23: Pass product tests # This is the commit message trinodb#24: Removed snowflake.properties in trino server dev # This is the commit message trinodb#25: Resolved issues 19 05 2022 and fixed tests # This is the commit message trinodb#26: Remove Types.VARBINARY # This is the commit message trinodb#27: Add private static SliceWriteFunction charWriteFunction # This is the commit message trinodb#28: Update test case # This is the commit message trinodb#29: Update plugin/trino-snowflake/src/main/java/io/trino/plugin/snowflake/SnowflakeClient.java Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> # This is the commit message trinodb#30: Update docs/src/main/sphinx/connector/snowflake.rst Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> # This is the commit message trinodb#31: Update plugin/trino-snowflake/pom.xml Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> # This is the commit message trinodb#32: Update plugin/trino-snowflake/src/main/java/io/trino/plugin/snowflake/SnowflakeClient.java Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> # This is the commit message trinodb#33: Resolved review open issues # This is the commit message trinodb#34: Disabled JDBC_TREAT_DECIMAL_AS_INT and fixed test case # This is the commit message trinodb#35: Updated properties file # This is the commit message trinodb#36: Updated properties files # This is the commit message trinodb#37: Renamed properties in Testing # This is the commit message trinodb#38: Revert "Renamed properties in Testing" This reverts commit 82f9eb3f3811e8d90a482f5359e98e7c729afa17. # This is the commit message trinodb#39: Renamed properties and fixed tests # This is the commit message trinodb#40: Update the way to pass ENV values for production test # This is the commit message trinodb#41: Update trino version to 388 # This is the commit message trinodb#42: Update Trino version 391 # This is the commit message trinodb#43: Update trino version to 394 # This is the commit message trinodb#44: Update to 395 # This is the commit message trinodb#45: Update to 411 # This is the commit message trinodb#46: Update and fix errors # This is the commit message trinodb#47: Build successfully with 411 # This is the commit message trinodb#48: Adding Bloomberg Snowflake connector. Fully tested with Trino 406. Untested with 410. Fix version number problem with build. Adding varchar type to mapping Snowflake. Adding --add-opens to enable support for Apache Arrow via shared memory buffers. Fixing some tests. Fix TestSnowflakeConnectorTest. TODO: testDataMappingSmokeTest: time and timestamp testSetColumnTypes: time and timestamp Fix type mapper Fix testconnector Remove unused argument from DeltaLakeMetastore.getTableLocation Extract removeS3Directory into DeltaLakeTestUtils Additionally, replace toUnmodifiableList with toImmutableList. Extract method in HiveMetastoreBackedDeltaLakeMetastore Don't install flush_metadata_cache procedure in Iceberg The procedure was unusable because Iceberg connector always disables caching metastore. Flush transaction log cache in Delta flush_metadata_cache procedure Co-Authored-By: Marius Grama <findinpath@gmail.com> Remove extra digest copy during Digest merge Reduce number of TDigestHistogram allocations On coordinator operators stats from all tasks will be merged. It does make sense to perform merging as bulk operation. Tune JDBC fetch-size automatically based on column count PostgreSQL, Redshift and Oracle connectors had hard-coded fetch-size value of 1000. The value was found not to be optimal when server is far (high latency) or when number of columns selected is low. This commit improves in the latter case by picking fetch size automatically based on number of columns projected. After the change, the fetch size will be automatically picked in the range 1000 to 100,000. Remove redundant LongDoubleState interface Fix import of wrong Preconditions class Test Iceberg cost-based plans with small files on TPC-DS Test against unpartitioned small Parquet files Upgrade Pinot libraries to 0.12.1 Simplify MaxDataSizeForStats and SumDataSizeForStats Block#getEstimatedDataSizeForStats is well defined for null positions. We can use this to replace NullableLongState with LongState. Fix formatting and simplify condition in HiveMetadata Skip listing Glue tables with invalid column types Exclude all synthetic columns in applyProjection validation DefaultJdbcMetadata#applyProjection already excludes the delete row id from validation. Same is now done for the merge row id column as well. Fail fast on unexpected case Fail, instead of returning, on an impossible case that was supposed to be handled earlier in a method. Remove redundant accessor calls Leverage information already had within the method. Remove Iceberg, Delta $data system table It was not intentional to expose a table's data as `a_table$data` "system" table. This commit removes support for these tables. Encapsulate table name class constructor Encapsulate constructors of IcebergTableName and DeltaLakeTableName. The classes are used primarily as utility classes. Constructor encapsulation is preparation to convert them into proper utility classes. Remove table name/type `from` parsing method After recent changes it became used only in tests. This also converts the IcebergTableName, DeltaLakeTableName into utility classes. Add more mirrors Future-proof against Ubuntu codename update This approach works as long the following assumptions hold: - the location of `/etc/os-release` file does NOT change - the name of the UBUNTU_CODENAME environment variable does NOT change - `eclipse-temurin` still uses Ubuntu as it's base Bring Docker build.sh help message in-line with reality Add retries in TestImpersonation Decrease number of old JDBC drivers tested Test parquet column mapping using name based and index based mapping Change t_char in AbstractTestHiveFileFormats to unpartitioned column Set thread name for DDL queries Before the change, the DDL tasks where being executed with a thread name of `dispatch-query-%d`. Fix thread names in TestEventDrivenTaskSource Remove unused constant in environment definition Capture column names in LocalQueryRunner Remove un-necessary projected call with assertThat in redshift connector Introduce new methods projected & exceptColumns take string varargs For better readability, replace projected(int... columns) with projected(String... columnNamesToInclude) and introduce exceptColumns(String... columnNamesToExclude) leveraging MaterializedResult.getColumnNames Access fields directly in DeltaLakeTableHandle.withProjectedColumns This makes it consistent with other methods like that, in particular with `DeltaLakeTableHandle.forOptimize`. Provide schema name in exception when Delta table lacks metadata Test Delta connector behavior for a corrupted table Handle corrupted Delta Lake tables with explicit table handle Previously, a corrupted table handle was represented with a `DeltaLakeTableHandle` missing a `MetadataEntry`. When drop support for corrupted tables was implemented, a connector could have only one table handle class. This commit improves the distinction by introducing a dedicated table handle class to represent corrupted tables. As a necessity, this class is explicitly handled in multiple `DeltaLakeMetadata` methods. This sets viable example to follow for implementing drop support for corrupted Iceberg tables as a follow-up. Note: `testGetInsertLayoutTableNotFound` test was removed, instead of being updated, since `ConnectorMetadata.getInsertLayout` cannot be reached for a corrupted table, as getting column list will fail earlier. Remove unnecessary override method in phoenix metadata Make queryModifier final Refactor testRenameTableToLongTableName to remove Iceberg override Refactor `testRenameTableToLongTableName` test so that Iceberg tests do not have to override the test method. Improve Iceberg test testCreateTableLikeForFormat code formatting Improve retries in AbstractTestHiveViews Previously we were retrying on "Execution Error, return code 2 from org.apache.hadoop.hive.ql.exec.mr.MapRedTask". As CI shown, the return code may vary, sometimes it is e.g. 1. In the meantime we introduced broader retry patterns for Hive query failures, so let's use these. Fix TestIcebergInsert.testIcebergConcurrentInsert timeout on CI The test seems dominated by `finishInsert` time. since bf04a72 `finishInsert` is slower as we commit twice (first committing data then statistics). Reduce lock contention in SqlStage Threads in the application were blocked on locks for a total of 4 h 19 min before this patch and 2 h 59 min after in concurrent benchmark with 40 nodes and 64 queries in parallel. Reduce lock contention in Query Use keySet in execution-query-purger Make setup of bucketed tables in HiveQueryRunner optional Bucketed tables are unnecessary in many tests Remove superfluous accumulator add function - Enhance the test case as well Add test for partitioned by non-lowercase column in Delta Fix failure when partition column contains uppercase in Iceberg Remove duplicate getParquetType method from ParquetPageSourceFactory Remove redundant boolean state from LongDecimalWithOverflowAndLongState Reduce synchronization on PipelinedStageExecution Removes synchronization from beginScheduling() and transitionToSchedulingSplits() both of which only perform state machine updates (if necessary) and do not require accessing synchronized state. Add NullablePosition to SumDataSizeForStats Avoids an extra null check as block.getEstimatedDataSizeForStats will also check for null Remove unused HiveHudiPartitionInfo.getTable method Co-Authored-By: Will Zhang <56012782+willzgw@users.noreply.github.com> Support DELETE statement in Ignite connector Add an example JDBC connector plugin Fix typo Split createDriverRunner method for partitioned and unpartitioned cases Use checkArgument formatting in StatementUtils Avoids an eager and unnecessary String.format call by letting checkArgument perform the required formatting only when the check fails. Avoid String.format in ExpressionFormatter Also replaces unnecessary usages of Guava's Joiner in favor of Collectors.joining where appropriate. Replace String.format with String concat Replaces simple String.format usages in non-exceptional code paths with simple string concatenations where applicable. Fix bad parameter count in code generator when column uses two slots Document hive.max-outstanding-splits-size property Add missing groups to testMigrateHiveBucketedOnMultipleColumns Remove unused updatedColumns from IcebergTableHandle Remove OkHttp as a runtime dependency for the engine Remove unused dependencies from discovery-server These are not used in embedded mode. Update to ASM 9.4 Update object storage definition in glossary Improve size accounting of SingleLongDecimalWithOverflowState Make classes in LongDecimalWithOverflowAndLongStateFactory private final Improve size accounting of SingleLongDecimalWithOverflowAndLongState Simplify DecimalAverageAggregation#inputShortDecimal Remove unsed NullableBooleanState Fix Kerberos ticket refresh The Hadoop UGI class handles ticket refresh only if the Subject is not provided externally. For external Subject UGI expects the refresh will be handled by the creator of the Subject which in our case we did not do. Because of this before this change any Trino query which ran longer than the ticket_lifetime failed with errors like GSS initiate failed [Caused by GSSException: No valid credentials provided (Mechanism level: Failed to find any Kerberos tgt)]. In Hadoop code the UGI instance also gets re-used in some places (e.g. DFSClient) which means we cannot just create a new UGI with refreshed credentials and return that since other parts of code will keep using the old UGI with expired credentials. So the fix is to create a new UGI, extract the credentials from it and update the existing UGI's credentials with them so that all users of the existing UGI also observe the new valid credentials. Extend the list of status codes retried in client This commit extends list of codes on which client will retry to: * HTTP_BAD_GATEWAY (502) * HTTP_UNAVAILABLE (503) * HTTP_GATEWAY_TIMEOUT(504) Allow listening for single container events Make environment listener always required Used enhanced switch Remove redundant local variable Remove redundant throws Use StringBuilder instead of string concatenation Fix typo in hive parquet doc Add test for trailing space in location in hive metadata Remove unnecessary and brittle tests Column names coming out of the query are not necessarily related to the column names in the table function. These tests are testing behavior that is not necessarily expected or guaranteed, so they are brittle and can break at any time. A couple of reasons why it's problematic: * Trino doesn't (yet) follow standard SQL identifier semantics. The column names might change between the output of the table function and the query output * At the query output all columns have names. Within the query they might not. A table function can produce an anonymous column, but the test will see "_col0". Upgrade Confluent version to 7.3.1 Updates transitive dependencies for Avro and ZooKeeper. Wire 4.x is required for Confluent 7.3.1 and is updated in the modules that need it, but leaves Wire at 3.x for the remaining modules. Fix potential Kerberos failure with SymlinkTextInputFormat Add benchmark for array filter object Optimize filter function performance with copyPositions Before the change: Benchmark (name) Mode Cnt Score Error Units BenchmarkArrayFilter.benchmark filter avgt 20 22.543 ± 0.979 ns/op BenchmarkArrayFilter.benchmarkObject filter avgt 20 42.045 ± 2.088 ns/op After the change: Benchmark (name) Mode Cnt Score Error Units BenchmarkArrayFilter.benchmark filter avgt 20 13.327 ± 0.359 ns/op BenchmarkArrayFilter.benchmarkObject filter avgt 20 34.443 ± 1.943 ns/op Add quantile_at_value function Co-authored-by: Peizhen Guo <pguo@fb.com> Use parent partitioning for aggregations If parent partitioning provides enough parallelism, and is a subset of the current node preferred partitioning (grouping keys for the aggregation) we can use the parent partitioning to skip data shuffle required by the parent. Extract MappedPageSource and MappedRecordSet to toolkit Introduce BaseJdbcConnectorTableHandle Extract methods in BaseJdbcClient These methods can be reused for Procedures PTF - Extract building columns from ResultSetMetaData as a separate method. - Extract creating connection based on session Add table function to execute stored procedure in SQLServer Use URI path for Glue location in tests Glue started throwing "InvalidInputException: One or more inputs failed validation" when getting a table if the table location doesn't have "file:" prefix in case of local file system. Test trino-main with JDK 20 Clarify comment in BigQuery ReadSessionCreator Consistently handle table types across BigQuery connector This also fixes a bug where createEmptyProjection failed for non-TABLE and non-VIEW even though those could be supported. Combine some redundant tests in BigQuery Remove duplicate test case Disable CSV quoting when quote character is zero Disable CSV escaping when escape character is zero Fix race condition in the hive table stats cache putIfAbsent method is not implemented in the EvictableCache because of race condition with invalidation so to avoid the race condition we use AtomicReference that at some cases can be thrown away, but it makes cached value fresh even if invalidation happens during value load Provide convenience overload to get MV storage table in test Setup global state before test methods `storageSchemaName` is defined on class level, so the storage schema should be created in `@BeforeClass`, not within a test. Allow Iceberg MV with partitioning transforms on timestamptz Allow creation of Iceberg Materialized Views partitioned with a temporal partitioning function on a `timestamp with time zone` column. In MVs, the `timestamp with time zone` columns are generally stored as text to preserve time zone information. However, this prevents use of temporal partitioning functions on these columns. The commit keeps `timestamp with time zone` columns with partitioning applied on them as `timestamp with time zone` in the storage table. An obvious downside to this approach is that the time zone information is erased and it is not known whether this aligns with user intention or not. A better solution would be to introduce a point-in-time type (trinodb#2273) to discern between the cases where time zone information is important (like Java's `ZonedDateTime`) from cases where only point-in-time matters (like Java's `Instant`). Remove backticks from backtick-unrelated test cases They were probably copied over from the preceding backtick test case. Reuse TrackingFileSystemFactory between connectors Move TrackingFileSystemFactory out of Iceberg tests to allow reuse e.g. with Delta Lake tests. Refactor Delta file operations tests to encapsulate checked code Pair tested operation and expected filesystem access counts in a single assertion calls. Similar to how it's done in `TestIcebergMetadataFileOperations`. Convert TestIcebergMetadataFileOperations helper to record Add Trino 411 release notes [maven-release-plugin] prepare release 411 [maven-release-plugin] prepare for next development iteration Enhance test for managed and external delta table location validation The purpose of deleting the transaction log directory is solely to confirm that when the DROP TABLE command is used, the table location is also removed when table is MANAGED TABLE. Improve naming of methods and fields to match Trino concepts Fix incorrect result when hidden directory exist in migrate procedure Add support for ADD COLUMN in Ignite connector Use a more specific name for all connectors smoke test suite Document the all connectors smoke test suite Verify pass-through specifications in tests Add more detailed check for TableFunctionProcessorNode.passThroughSpecifications in TableFunctionProcessorMatcher. Prune unreferenced pass-through columns of table function Verify required columns in tests Add check for TableFunctionProcessorNode.requiredSymbols in TableFunctionProcessorMatcher. Verify hashSymbol in tests Add check for TableFunctionProcessorNode.hashSymbol in TableFunctionProcessorMatcher. Prune unreferenced columns of table function source Test table function column pruning in query plan Remove table function with empty source Adds an optimizer rule to remove TableFunctionProcessorNode with source being an empty relation, based on the "prune when empty" property. Test pruning of redundant table function in query plan Test table functions with column and node pruning optimizations Fix typo in TestDeltaLakePerTransactionMetastoreCache Extract assertMetastoreInvocations method Use CountingAccessHiveMetastore in TestDeltaLakePerTransactionMetastoreCache Make cleanup methods alwaysRun This attribute says that this after-method will get executed even if the methods executed previously failed or were skipped. Note that it also applies to skipped test methods, so if the tests were skipped for some reason, the cleanup won't run. This attribute will ensure that the clean-up will run even in this case. Failure to run clean up may cause secondary effects, especially our resource leak detector; failure on this will in turn mask other errors, the ones which caused the tests to be skipped in the first place. Add a check to enforce alwaysRun = true on test after-methods See the previous commit for details. This check will enforce that the `alwaysRun = true` is present. Remove redundant call toString() Remove use of deprecated isEqualToComparingFieldByFieldRecursively Use usingRecursiveComparison instead deprecated isEqualToComparingFieldByFieldRecursively Remove unused helper methods in delta-lake connector Add an explict config to define standardSplitSizeInBytes in FTE Implement adaptive task sizing for arbitrary distribution in FTE Improve task sizing for hash distribution in FTE Round up targetPartitionSizeInBytes to a multiple of minTargetPartitionSizeInBytes For adaptive task sizing in ArbitraryDistributionSplitAssigner Adjust retry policy for dropping delta tables backed by AWS Glue on Databricks Fix output rendering issue in docs Alphabetize glossary entries Add use_cost_based_partitioning Use use_cost_based_partitioning instead of use_exact_partitioning to control the cost based optimization to prefer parent partitioning. The motivation is to be able to disable the optimization if the NDV statistics are overestimated and the optimization would hurt parallelism. Provide injection mechanism for the file system factory Reorder instance and static fields in FlushMetadataCacheProcedure Flush extended statistics in Delta's flush_metadata_cache() Clean up Delta code a bit Test Delta Lake query file system accesses Ensure that TestingHydraIdentityProvider is releasing resources Update maven to 3.9.1 Expose rule stats in QueryStats Expose optimizer rule statistics per query in QueryInfo JSON. The number of rules exposed could be adjusted using the `query.reported-rule-stats-limit` configuration parameter. Cleanup BigintGroupByHash instanceof checks Include QueryId in assertDistrubutedQuery failure message Remove TestClickHouseConnectorTest There are 4 smoke tests and 2 connectors test. Remove TestClickHouseConnectorTest as a redundant test. Remove base class for ClickHouse connector test Run smoke test for BigQuery arrow serialization We want to verify SELECT behavior for Arrow serialization in BigQuery. Smoke test and the existing type mapping test should be enough. Make construction parameters final Support arithmetic predicate pushdown for Phoenix Make LeafTableFunctionOperator handle EmptySplit correctly Remove CatalogHandle from TableFunctionProcessorNode Exclude snakeyaml from elasticsearch dependencies It's a transitive dependency of elasticsearch-x-content, which we use in ElasticsearchLoader to load tpch data to Elasticsearch with json encoding. Yaml support is not needed at all. Pass partition values as optional to DeltaLakePageSource The partition values list is filled only when row ID column is projected, so it's a conditional information. When row ID is not present, pass it as the empty optional, rather than list that happens to be empty. Add cleaner FixedPageSource constructors Previously, the only constructor would take `Iterable`, which is nice, but it would also materialize it twice (once in the constructor to calculate memory usage). The commit adds a constructor taking a `List` (so double iteration is not a problem) and one taking `Iterator` and delivering on the promise to iterate once. The old constructor is kept deprecated, but apparently all usages use the new, list-based constructor. Project a data column in MinIO access test Read a data column to ensure the data file gets read. This increases number of accesses to a file, because both footer and data are being read. Accelerate Delta when reading partition columns only Regenerate expected test plans with one-click Traverse into JsonObject members in AST Before this change, JsonObject members were not visited in AstVisitor. As a result, aggregations or parameters inside the members were not supported. Traverse into JsonArray elements in AST Before this change, JsonArray elements were not visited in AstVisitor. As a result, aggregations or parameters inside the elements were not supported. Document avro.schema.literal property use for interpreting table data Update Oracle JDBC driver version to 21.9.0.0 Document predicate pushdown support for string-type columns in SQL Server Enable oracle.remarks-reporting.enabled in connector test Remove unnecessary wrapping of IOException in TransactionLogTail Translate `The specified key does not exist` to FileNotFoundException Relax test assertion It is possible for more than one task to fail due to injected failure. Remove obsolete assertion Lowercase bucketing and sort column names In the metastore, the bucketing and sorting column names can differ in case from its corresponding table column names. This change makes certain that, even though a table can be delivered by the metastore with such inconsistencies, Trino will lowercase the same bucketing and sort column names to ensure they correspond to the data column names. Add test for RenameColumnTask Migrate assertStatement in TestSqlParser.testRenameColumn Allow configuring a custom DNS resolver for the JDBC driver Reorganize Hudi connector documentation Migrate some assertExpression in TestSqlParser Look for a non-Trino protocol without using X-User-* Update ASM to 9.5 Reorganize Iceberg connector documentation Reorganize Delta Lake connector documentation Fix handling of Hive ACID tables with hidden directories Test CREATE TABLE AS SELECT in Ignite type mapping Additionally, check time zones in setUp method. Override equals and hashCode in Delta Identifier Change Object to Row in testCheckConstraintCompatibility Support arithmetic binary in Delta check constraints Introduce EmptyTableFunctionHandle as default handle Before this change, if a table function did not pass a ConnectorTableFunctionHandle in the TableFunctionAnalysis, the default handle was used, which was an anonymous implementation of ConnectorTableFunctionHandle. It did not work with table functions executed by operator, due to lack of serialization. This change introduces EmptyTableFunctionHandle, and sets it as default. Support returning anonymous columns by table functions Per SQL standard, all columns must be named. In Trino, we support unnamed columns. This change adjusts table functions so that they can return anonymous columns. It is achieved by changing the Descriptor structure so that field name is optional. This optionality can only be used for the returned type of table functions. Descriptor arguments pased by the user as well as default values for descriptor argumens have mandatory field names. Add table function `exclude_columns` Bump spotbugs-annotations version Airbase already has `4.7.3` Remove ValidateLimitWithPresortedInput It's not powerful enough to validate properties of plans that get modified by predicate pushdown after AddExchanges runs, resulting in false positives such as trinodb#16768 Use OrcReader#MAX_BATCH_SIZE = 8 * 1024 Previous value 8196 was bigger than PageProcessor#MAX_BATCH_SIZE causing PageProcessor to create small, 4 position pages every other page. Bring DistinguishedNameParser from okhttp3 Inline OkHostnameVerifier and Util.verifyAsIpAddress They were removed in OkHttp 4.x and we still rely on the legacy SSL hostname verification Update okhttp to 4.10.0 Fix testTableWithNonNullableColumns to update NOT NULL column in Delta Add for support creating table with comment for more Jdbc based connectors (trinodb#16135) Allow PostgreSQL create table with comment Also allow to set a comment for PostgreSQL tables Support sum(distinct) for jdbc connectors Add Trino 412 release notes [maven-release-plugin] prepare release 412 [maven-release-plugin] prepare for next development iteration Add doc for ignite join pushdown Add missing config properties to Hive docs Co-Authored-By: Marius Grama <findinpath@gmail.com> Fix layout in Iceberg documentation Add docs for property to skip glue archive Fix typo Support nested timestamp with time zone in Delta Lake Add test for duplicated partition statistics on Thrift metastore Support table comment for oracle connector Allow Oracle create table with comment Also allow to set table comment for Oracle tables Support MERGE for Phoenix connector Remove unused class FallbackToFullNodePartitionMemoryEstimator. Remove unnecessary dependency management in trino-pinot The `protobuf-java` one was overriding a corresponding declaration in the parent POM, and was effectively downgrading it. The other two were not used at all. Bump Protobuf version Add Snowflake JDBC Connector Update trino snapshot version to 372 Update format of the doc of snowflake Update trino jdbc library import Fix date formatter from yyyy to uuuu Fix date test case Remove defunct property allow-drop-table Update trino version to 374 Update snowflake config to adapt 374 Update the range of the test of Date type Update to version 375 Fix snowflake after updating to 375 Update to 381 Fix mvn pom import Format snowflake.rst Reorderd Data tests in type mapping Update function code Add product test Rename product test tablename Add Env, Suite and properties of Snowflake for production test Add trinoCreateAndInsert() Refactor snowflake from single node to multi node Pass product tests Removed snowflake.properties in trino server dev Resolved issues 19 05 2022 and fixed tests Remove Types.VARBINARY Add private static SliceWriteFunction charWriteFunction Update test case Update plugin/trino-snowflake/src/main/java/io/trino/plugin/snowflake/SnowflakeClient.java Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> Update docs/src/main/sphinx/connector/snowflake.rst Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> Update plugin/trino-snowflake/pom.xml Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> Update plugin/trino-snowflake/src/main/java/io/trino/plugin/snowflake/SnowflakeClient.java Co-authored-by: Yuya Ebihara <ebyhry@gmail.com> Resolved review open issues Disabled JDBC_TREAT_DECIMAL_AS_INT and fixed test case Updated properties file Updated properties files Renamed properties in Testing Revert "Renamed properties in Testing" This reverts commit 82f9eb3f3811e8d90a482f5359e98e7c729afa17. Renamed properties and fixed tests Update the way to pass ENV values for production test Update trino version to 388 Update Trino version 391 Update trino version to 394 Update to 395 Update to 411 Update and fix errors Build successfully with 411 Adding Bloomberg Snowflake connector. Fully tested with Trino 406. Untested with 410. Adding varchar type to mapping Snowflake. Adding --add-opens to enable support for Apache Arrow via shared memory buffers. Fixing some tests. Fix TestSnowflakeConnectorTest. TODO: testDataMappingSmokeTest: time and timestamp testSetColumnTypes: time and timestamp Fix type mapper Fix testconnector Update version to 413-SNAPSHOT Added support for HTTP_PROXY testing. Connector doesnt support setColumnType. This causes lots of problems with Snowflake server. Disabled the testsetColumnTypes testing. Fixed and skipped error tests

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

Various style fixes and cleanup (#15) (#17) Co-authored-by: Martin Traverso <mtraverso@gmail.com> Various style fixes and cleanup (#15) Update the github CI (#12) * Add Snowflake JDBC Connector * Add snowflake in the ci Add Snowflake JDBC Connector (#11) Had to redo the connector because all the rebases caused havoc

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

Various style fixes and cleanup (#15) (#17) Co-authored-by: Martin Traverso <mtraverso@gmail.com> Various style fixes and cleanup (#15) Update the github CI (#12) * Add Snowflake JDBC Connector * Add snowflake in the ci Add Snowflake JDBC Connector (#11) Had to redo the connector because all the rebases caused havoc

***** Update ci and delete no use lines (#20) Approved Update according to reviews 11/01/2024 Various style fixes and cleanup (#15) (#17) Co-authored-by: Martin Traverso <mtraverso@gmail.com> Various style fixes and cleanup (#15) Update the github CI (#12) * Add Snowflake JDBC Connector * Add snowflake in the ci Add Snowflake JDBC Connector (#11) Had to redo the connector because all the rebases caused havoc (cherry picked from commit dafaa5d)

***** Update ci and delete no use lines (trinodb#20) Approved Update according to reviews 11/01/2024 Various style fixes and cleanup (trinodb#15) (trinodb#17) Co-authored-by: Martin Traverso <mtraverso@gmail.com> Various style fixes and cleanup (trinodb#15) Update the github CI (trinodb#12) * Add Snowflake JDBC Connector * Add snowflake in the ci Add Snowflake JDBC Connector (trinodb#11) Had to redo the connector because all the rebases caused havoc

***** Update ci and delete no use lines (#20) Approved Update according to reviews 11/01/2024 Various style fixes and cleanup (#15) (#17) Co-authored-by: Martin Traverso <mtraverso@gmail.com> Various style fixes and cleanup (#15) Update the github CI (#12) * Add Snowflake JDBC Connector * Add snowflake in the ci Add Snowflake JDBC Connector (#11) Had to redo the connector because all the rebases caused havoc

prrvchr · 2024-05-31T11:14:45Z

The workaround is to only create tables in lower case but do you have anything planned on your side to resolve this problem?

shantanu-dahiya · 2024-09-19T19:11:16Z

Does the contributing team have plans to pick this up anytime soon?

martint added the enhancement New feature or request label Jan 21, 2019

martint assigned Praveen2112 Jan 21, 2019

martint mentioned this issue Mar 4, 2019

Record if authorization identifier is delimited or not #313

Open

kokosing mentioned this issue Mar 8, 2019

[WIP] : Add support Case sensitive identifiers #354

Closed

dain added the roadmap Top level issues for major efforts in the project label Mar 13, 2019

findepi mentioned this issue May 31, 2019

Phoenix Connector Lowercase Table/View could not be find #858

Closed

findepi mentioned this issue Jul 9, 2019

MongoDB collection name case-sensitivity #1102

Closed

Praveen2112 mentioned this issue Aug 26, 2019

JdbcMetadata#listTableColumns is broken when there are two tables which differ by letter casing #1373

Closed

Praveen2112 mentioned this issue Sep 29, 2019

Add support Case sensitive identifiers #1302

Closed

Praveen2112 mentioned this issue Dec 26, 2019

Add support for case sensitive identifiers #2350

Closed

tooptoop4 mentioned this issue Jun 6, 2020

Casing of URLs getting changed to lower case (can't get data from URLs containing upper case letters) snowlift/trino-storage#18

Closed

This was referenced Jun 9, 2020

MySQL case-sensitive table names not working #3974

Closed

Redis tableName should be case-sensitive #3999

Open

findepi mentioned this issue Jun 29, 2020

Quote column name in Phoenix CREATE TABLE statement #3601

Merged

This was referenced Aug 16, 2020

Query failure for information_schema.columns when duplicated table name exists #4853

Open

MySQL Connector has issue with Pascal case Database name #4930

Closed

kokosing mentioned this issue Aug 25, 2020

Fix bug when querying system.jdbc.tables #4965

Closed

kokosing mentioned this issue Feb 2, 2021

Trino + Pinot Connector - Lowercasing table name in query #6789

Closed

hashhar mentioned this issue Mar 14, 2021

trino table column is case Insensetive for elasticsearch connector #7276

Closed

findepi mentioned this issue Mar 15, 2021

How to make trino table name case sensetive? #7286

Closed

ebyhr mentioned this issue Apr 24, 2021

Database names with a capital letter. Clickhouse 21.1.2.15. #7702

Closed

jtcohen6 mentioned this issue May 5, 2021

Updates catalog queries to use strict equality dbt-labs/dbt-presto#40

Merged

kasiafi mentioned this issue Apr 20, 2023

Support adding a field with ADD COLUMN in Iceberg #16321

Merged

This was referenced Aug 1, 2023

Fix bug Delta column operations erases its properties on uppercase names #18123

Merged

Support setting a field type with SET DATA TYPE statement in engine and Iceberg #18395

Merged

ebyhr mentioned this issue Sep 5, 2023

mongoDb : SELECT query fails with "Multiple entries with same key" #18925

Closed

This was referenced Dec 13, 2023

Add support for Elasticsearch query string syntax #1662

Merged

Query throws exception on mongodb connector if database have duplicate collection name as case insensitive #20167

Closed

yuuteng added a commit to yuuteng/trino that referenced this issue Jan 11, 2024

Various style fixes and cleanup (trinodb#15) (trinodb#17)

fe244d3

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

yuuteng added a commit to yuuteng/trino that referenced this issue Jan 11, 2024

Various style fixes and cleanup (trinodb#15) (trinodb#17)

69d4ab2

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

ebyhr pushed a commit that referenced this issue Jan 14, 2024

Various style fixes and cleanup (#15) (#17)

e6dd1fd

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

ebyhr pushed a commit that referenced this issue Jan 14, 2024

Various style fixes and cleanup (#15) (#17)

9bdf104

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

hashhar mentioned this issue Jan 19, 2024

File based column mapping for JDBC connectors #20413

Closed

yuuteng added a commit to yuuteng/trino that referenced this issue Feb 22, 2024

Various style fixes and cleanup (trinodb#15) (trinodb#17)

fd1ef29

Co-authored-by: Martin Traverso <mtraverso@gmail.com>

ebyhr mentioned this issue May 31, 2024

DatabaseMetadata.getTables() is sensitive to the table name when table creation is not #22204

Closed

findinpath mentioned this issue Jun 4, 2024

Add Parquet bloom filter write support to Iceberg connector #21602

Merged

ebyhr mentioned this issue Jul 16, 2024

Support a case-sensitive column name for Apache Iceberg Rest Catalog #22677

Open

findinpath mentioned this issue Aug 15, 2024

Fix handling of duplicate column names in parquet reader #23050

Merged

ebyhr mentioned this issue Oct 21, 2024

Cannot query postgres when tables have complex names/case #23820

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for case sensitive identifiers #17

Add support for case sensitive identifiers #17

martint commented Jan 21, 2019 •

edited

Loading

martint commented Mar 4, 2019 •

edited

Loading

findepi commented Mar 5, 2019

martint commented Mar 5, 2019

findepi commented Apr 23, 2019

martint commented Apr 23, 2019

SQL_IDENTIFIER domain

findepi commented Apr 23, 2019

Praveen2112 commented Apr 24, 2019

kokosing commented Sep 2, 2020

prrvchr commented May 31, 2024

shantanu-dahiya commented Sep 19, 2024

Add support for case sensitive identifiers #17

Add support for case sensitive identifiers #17

Comments

martint commented Jan 21, 2019 • edited Loading

martint commented Mar 4, 2019 • edited Loading

findepi commented Mar 5, 2019

martint commented Mar 5, 2019

findepi commented Apr 23, 2019

Proposed representation

martint commented Apr 23, 2019

SQL_IDENTIFIER domain

findepi commented Apr 23, 2019

Praveen2112 commented Apr 24, 2019

kokosing commented Sep 2, 2020

prrvchr commented May 31, 2024

shantanu-dahiya commented Sep 19, 2024

martint commented Jan 21, 2019 •

edited

Loading

martint commented Mar 4, 2019 •

edited

Loading