[jira] [Commented] (DRILL-5311) C++ connector connect doesn't check handshake result for timeout

["Laurent Goujon (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/DRILL-5311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15894920#comment-15894920 ] 

Laurent Goujon commented on DRILL-5311:
---------------------------------------

Did some testing using querySubmitter:

before:
{noformat}
$ ./querySubmitter type=server connectStr=local=localhost:31010 api=meta logLevel=trace
Connector:
	name:Apache Drill C++ client
	version:1.10.0

Server:
	name:
	version:

Metadata:
	all tables are selectable: 0
	catalog separator: .
	catalog term: catalog
	COLLATE support: 0
	correlation names: 3
	date time functions: CURDATE, CURTIME, NOW, QUARTER
	date time literals support: 65278
	GROUP BY support: 0
	identifier case: 1
	identifier quote string: `
	max binary literal length: 0
	max catalog name length: 1024
	max char literal length: 0
	max column name length: 1024
	max columns in GROUP BY: 1024
	max columns in ORDER BY: 0
	max columns in SELECT: 0
	max cursor name length: 1024
	max logical lob size: 0
	max row size: 0
	max schema name length: 1024
	max statement length: 0
	max statements: 0
	max table name length: 1024
	max tables in SELECT: 0
	max user name length: 1024
	NULL collation: 1
	numeric functions: ABS, EXP, LOG, LOG10, MOD, POWER
	OUTER JOIN support: 14
	quoted identifier case: 1
	SQL keywords: ABS,ALLOW,ARRAY,ASENSITIVE,ASYMMETRIC,ATOMIC,BIGINT,BINARY,BLOB,BOOLEAN,CALL,CALLED,CARDINALITY,CEIL,CEILING,CLOB,COLLECT,CONDITION,CORR,COVAR_POP,COVAR_SAMP,CUBE,CUME_DIST,CURRENT_CATALOG,CURRENT_DEFAULT_TRANSFORM_GROUP,CURRENT_PATH,CURRENT_ROLE,CURRENT_SCHEMA,CURRENT_TRANSFORM_GROUP_FOR_TYPE,CYCLE,DATABASE,DATABASES,DENSE_RANK,DEREF,DETERMINISTIC,DISALLOW,DYNAMIC,EACH,ELEMENT,EVERY,EXP,EXPLAIN,EXTEND,FILES,FILTER,FIRST_VALUE,FLOOR,FREE,FUNCTION,FUSION,GROUPING,HOLD,IF,IMPORT,INOUT,INTERSECTION,LARGE,LAST_VALUE,LATERAL,LIMIT,LN,LOCALTIME,LOCALTIMESTAMP,MEMBER,MERGE,METADATA,METHOD,MOD,MODIFIES,MULTISET,NCLOB,NEW,NONE,NORMALIZE,OFFSET,OLD,OUT,OVER,OVERLAY,PARAMETER,PARTITION,PERCENTILE_CONT,PERCENTILE_DISC,PERCENT_RANK,POWER,RANGE,RANK,READS,RECURSIVE,REF,REFERENCING,REFRESH,REGR_AVGX,REGR_AVGY,REGR_COUNT,REGR_INTERCEPT,REGR_R2,REGR_SLOPE,REGR_SXX,REGR_SXY,REGR_SYY,RELEASE,REPLACE,RESET,RESULT,RETURN,RETURNS,ROLLUP,ROW,ROW_NUMBER,SAVEPOINT,SCHEMAS,SCOPE,SEARCH,SENSITIVE,SHOW,SIMILAR,SPECIFIC,SPECIFICTYPE,SQLEXCEPTION,SQLWARNING,SQRT,START,STATIC,STDDEV_POP,STDDEV_SAMP,STREAM,SUBMULTISET,SYMMETRIC,SYSTEM,TABLES,TABLESAMPLE,TINYINT,TREAT,TRIGGER,UESCAPE,UNNEST,UPSERT,USE,VARBINARY,VAR_POP,VAR_SAMP,WIDTH_BUCKET,WINDOW,WITHIN,WITHOUT
	schema term: schema
	search escape string: \
	special characters: 
	string functions: CONCAT,INSERT,LCASE,LENGTH,LOCATE,LTRIM,RTRIM,SUBSTRING,UCASE
	sub query support: 46
	system functions: 
	table term: table
	UNION support: 6
	BLOB included in max row size: 1
	catalog at start: 1
	column aliasing supported: 1
	LIKE escape clause supported: 1
	NULL plus non NULL equals to NULL: 1
	read-only: 0
	SELECT FOR UPDATE supported: 0
	transaction supported: 0
	unrelated columns in ORDER BY supported: 1
{noformat}

Note that how the connection didn't fail and it is still possible to query API (although results are not populated correctly).

after:
{noformat}
$ ./querySubmitter type=server connectStr=local=localhost:31010 api=meta logLevel=trace
Failed to connect with error: [30015]Handshake Timeout. (Using:local=localhost:31010)
{noformat}


> C++ connector connect doesn't check handshake result for timeout
> ----------------------------------------------------------------
>
>                 Key: DRILL-5311
>                 URL: https://issues.apache.org/jira/browse/DRILL-5311
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Client - C++
>            Reporter: Laurent Goujon
>            Assignee: Sudheesh Katkam
>              Labels: ready-to-commit
>
> The C++ connector connect methods returns okay as soon as the tcp connection is succesfully established between client and server, and the handshake message is sent. However it doesn't wait for handshake to have completed.
> The consequence is that if handshake failed, the error is deferred to the first query, which might be unexpected by the application.
> I believe that validateHanshake method in drillClientImpl should wait for the handshake to complete, as it seems a bit more saner...



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)