[SPARK-22122][SQL] Use analyzed logical plans to count input rows in TPCDSQueryBenchmark #19344
Conversation
|
Test build #82167 has finished for PR 19344 at commit
|
|
@gatorsmile if you get time, please check this. thanks. |
|
ping |
There was a problem hiding this comment.
Why not using the plan that has been analyzed?
There was a problem hiding this comment.
The analyzer rule CTESubstitution will replace With
There was a problem hiding this comment.
oh, yea. Since the original code does so, I just added the logic. But, the suggestion sounds good to me, so I'll update soon. Thanks.
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
3 times, most recently
from
0df2663 to
dd84919
Compare
There was a problem hiding this comment.
If the plan is successfully analyzed, UnresolvedRelation should not exist
|
Test build #82280 has finished for PR 19344 at commit
|
|
Test build #82281 has finished for PR 19344 at commit
|
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
2 times, most recently
from
3f31048 to
f7359af
Compare
|
@gatorsmile ok, fixed. Also, I checked this code could collect all the relations. |
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
from
f7359af to
489f2a2
Compare
maropu
changed the title
|
Test build #82301 has finished for PR 19344 at commit
|
|
Test build #82303 has finished for PR 19344 at commit
|
| case _ => | ||
| } | ||
| } | ||
| spark.sql(queryString).queryExecution.analyzed.map { |
| } | ||
| spark.sql(queryString).queryExecution.analyzed.map { | ||
| case SubqueryAlias(name, _: LogicalRelation) => | ||
| queryRelations.add(name) |
There was a problem hiding this comment.
Add another case for HiveTableRelation
There was a problem hiding this comment.
IIUC ditto; HiveTableRelation never happens here.
| } | ||
| } | ||
| spark.sql(queryString).queryExecution.analyzed.map { | ||
| case SubqueryAlias(name, _: LogicalRelation) => |
There was a problem hiding this comment.
Why not using LogicalRelation 's catalogTable? Just issue an exception if it is None. I think this benchmark will not hit None
There was a problem hiding this comment.
I checked again and I found we can't use catalogTable here because these TPCDS tables are locally temporary ones (IIUC these tables are always transformed into ScalaAlias(LocalRelation)).
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
from
6dfb004 to
00cfb21
Compare
|
Test build #82309 has finished for PR 19344 at commit
|
|
Test build #82311 has finished for PR 19344 at commit
|
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
from
00cfb21 to
5691cf6
Compare
|
Test build #82314 has finished for PR 19344 at commit
|
|
retest this please |
| } | ||
| } | ||
| spark.sql(queryString).queryExecution.analyzed.foreach { | ||
| case SubqueryAlias(name, _: LogicalRelation) => |
There was a problem hiding this comment.
Could we add all the three scenarios, although the current codes only use temp views?
|
Test build #82329 has finished for PR 19344 at commit
|
|
Test build #82340 has finished for PR 19344 at commit
|
maropu
force-pushed
the
RespectWithInTPCDSBench
branch
from
cbac959 to
8d8a9ff
Compare
|
LGTM pending Jenkins |
|
Test build #82341 has finished for PR 19344 at commit
|
|
btw, could we also add |
|
These modified test cases are not following the standards. Impala added extra (partition) predicates. The perf results are misleading. |
|
Thanks! Merged to master. |
|
ok, thanks! |
3 participants
What changes were proposed in this pull request?
Since the current code ignores WITH clauses to check input relations in TPCDS queries, this leads to inaccurate per-row processing time for benchmark results. For example, in
q2, this fix could catch all the input relations:web_sales,date_dim, andcatalog_sales(the current code catchesdate_dimonly). The one-third of the TPCDS queries uses WITH clauses, so I think it is worth fixing this.How was this patch tested?
Manually checked.