Please help tune this query

Question

Following is the sql code
SELECT mbr_contr_med_cov_flag,
           mbr_contr_rx_cov_flag,
           src_mbr_id,
           cov_cd_eff_dt,
           MIN (
              TO_DATE (
                 SYS_CONNECT_BY_PATH (
                    DECODE (LEVEL, 1, TO_CHAR (elig_eff_dt), ''),
              AS beg_dt,
           MAX (elig_term_dt) term_dt
      FROM temp_elig_mem
CONNECT BY PRIOR src_mbr_id = src_mbr_id
           AND PRIOR elig_term_dt + 1 = elig_eff_dt
  GROUP BY mbr_contr_med_cov_flag,
           mbr_contr_rx_cov_flag,
           src_mbr_id,
           cov_cd_eff_dtExplain plan:
PLAN_TABLE_OUTPUT
| Id  | Operation                     | Name          | Rows  | Bytes | Cost  |
|   0 | SELECT STATEMENT              |               |   311K|    13M|   426 |
|   1 |  HASH GROUP BY                |               |   311K|    13M|   426 |
|   2 |   CONNECT BY WITHOUT FILTERING|               |       |       |       |
|   3 |    TABLE ACCESS FULL          | TEMP_ELIG_MEM |   311K|    13M|   411 |
-------------------------------------------------------------------------------Funny part is if I remove a 'PRIOR' from the code it runs faster. I mean the following code with one 'PRIOR' taken out runs faster:
SELECT mbr_contr_med_cov_flag,
           mbr_contr_rx_cov_flag,
           src_mbr_id,
           cov_cd_eff_dt,
           MIN (
              TO_DATE (
                 SYS_CONNECT_BY_PATH (
                    DECODE (LEVEL, 1, TO_CHAR (elig_eff_dt), ''),
              AS beg_dt,
           MAX (elig_term_dt) term_dt
      FROM temp_elig_mem
CONNECT BY PRIOR src_mbr_id = src_mbr_id
           AND elig_term_dt + 1 = elig_eff_dt
  GROUP BY mbr_contr_med_cov_flag,
           mbr_contr_rx_cov_flag,
           src_mbr_id,
           cov_cd_eff_dt

Answer 1

user2361373 wrote:
The problem may be due to MIN and MAX().Whole query is a mess. Just look at decode. For LEVEL= 1 it returns TO_CHAR(elig_eff_dt). For other levels it returns NULL. Therefore:
                    SYS_CONNECT_BY_PATH(
                                        DECODE(LEVEL,1,TO_CHAR(elig_eff_dt),''),
                                       )returns TO_CHAR(elig_eff_dt) of LEVEL = 1 followed by bunch of spaces. Then OP converts it back to DATE. In other words, regardless what LEVEL is it returns top level (LEVEL = 1) elig_eff_dt, which is nothing but CONNECT_BY_ROOT elig_eff_dt. So query can be simplified:
SELECT  mbr_contr_med_cov_flag,
        mbr_contr_rx_cov_flag,
        src_mbr_id,
        cov_cd_eff_dt,
        MIN(CONNECT_BY_ROOT elig_eff_dt) AS beg_dt,
        MAX(elig_term_dt) term_dt
  FROM temp_elig_mem
  CONNECT BY PRIOR src_mbr_id = src_mbr_id
         AND PRIOR elig_term_dt + 1 = elig_eff_dt
  GROUP BY mbr_contr_med_cov_flag,
           mbr_contr_rx_cov_flag,
           src_mbr_id,
           cov_cd_eff_dt
/Now:
  CONNECT BY PRIOR src_mbr_id = src_mbr_id
         AND elig_term_dt + 1 = elig_eff_dtIt obviously lacks PRIOR:
  CONNECT BY PRIOR src_mbr_id = src_mbr_id
         AND PRIOR elig_term_dt + 1 = elig_eff_dtNow, assuming table is relatively big, since there is no START WITH I suspect number of unnecessary rows built by hierarchy is a big part of performance issue. I'd add:
START WITH (src_mbr_id,elig_eff_dt) IN (SELECT src_mbr_id,MIN(elig_eff_dt) FROM temp_elig_mem GROUP BY src_mbr_id)And if all rows in each hierarchy have same mbr_contr_med_cov_flag, mbr_contr_rx_cov_flag, cov_cd_eff_dt:
MAX(elig_term_dt) term_dtis nothing but elig_term_dt from the hierarchy leaf row. Then whole query can be simplified to:
SELECT  mbr_contr_med_cov_flag,
        mbr_contr_rx_cov_flag,
        src_mbr_id,
        cov_cd_eff_dt,
        CONNECT_BY_ROOT elig_eff_dt AS beg_dt,
        elig_term_dt term_dt
  WHERE CONNECT_BY_ISLEAF = 1
  FROM temp_elig_mem
  START WITH (src_mbr_id,elig_eff_dt) IN (SELECT src_mbr_id,MIN(elig_eff_dt) FROM temp_elig_mem GROUP BY src_mbr_id)
  CONNECT BY PRIOR src_mbr_id = src_mbr_id
         AND PRIOR elig_term_dt + 1 = elig_eff_dt
/SY.
Edited by: Solomon Yakobson on Jan 16, 2012 7:08 PM