Re: Passing query string to workers

On Mon, Jan 23, 2017 at 2:46 PM, Kuntal Ghosh
<kuntalghosh(dot)2007(at)gmail(dot)com> wrote:
> I've looked into the patch. I've some comments regarding that.
>
> +#define PARALLEL_KEY_QUERY_TEXT UINT64CONST(0xE000000000000010)
> It should be UINT64CONST(0xE00000000000000A)
>
> + query_len = strlen(query_data) + 1;
> + shm_toc_estimate_chunk(&pcxt->estimator, query_len);
> + shm_toc_estimate_keys(&pcxt->estimator, 1);
> +
> Adding a comment before this will be helpful. You should maintain the
> same order for estimating and storing the query string. For example,
> as you've estimated the space for query string after estimation of dsa
> space, you should store the same after creating dsa. Besides, I think
> the appropriate place for this would be before planned statement.
Fixed.

> The regression test for select_parallel is failing after applying the
> patch. You can reproduce the scenario by the following sql statements:
>
> CREATE TABLE t1(a int);
> INSERT INTO t1 VALUES (generate_series(1,100000));
> SET parallel_setup_cost=0;
> SET parallel_tuple_cost=0;
> SET min_parallel_relation_size=0;
> SET max_parallel_workers_per_gather=4;
> SELECT count(*) FROM t1;
>
> It is showing the following warning.
> WARNING: problem in alloc set ExecutorState: detected write past
> chunk end in block 0x14f5310, chunk 0x14f6c50
Fixed.

Thanks a lot Kuntal for the review, please find attached patch for the
revised version.
--
Regards,
Rafia Sabih
EnterpriseDB: http://www.enterprisedb.com/