SQL Upsert using load periodically not saving all rows
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Trafodion |
New
|
High
|
Suresh Subbiah |
Bug Description
When running a script that initiates 32 parallel streams loading a table, we have found that periodically there are gaps in the resulting saved data...for example we will find that we are missing stock items #29485 thru #30847 inclusive for Warehouse #5. The number of gaps found for a given load run varies...normally none, but I've seen as many as eight gaps of missing data.
The sql statement used in all streams is as follows:
sql_statement = "upsert using load into " + stock_table_name
+ " (S_I_ID, S_W_ID, S_QUANTITY, S_DIST_01, S_DIST_02, S_DIST_03, S_DIST_04,"
+ " S_DIST_05, S_DIST_06, S_DIST_07, S_DIST_08, S_DIST_09, S_DIST_10,"
+ " S_YTD, S_ORDER_CNT, S_REMOTE_CNT, S_DATA)"
+ " values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)";
This is not easily repeatable…I’ve run the script to drop/create/load this table 12 times today, resulting in some missing rows 4 of the 12 times. Worst case we were missing 0.03% of the required rows in the table…obviously, ANY missing data is not acceptable.
Our test environment control parameters (in case any are of value to you)...
OrderEntryLoader
Load Starting : 2015-02-13 04:58:13
PropertyFile : trafodion.
Datebase : trafodion
Schema : trafodion.javabench
ScaleFactor : 512
Streams : 32
Maintian : true
Load : true
AutoCommit : true
BatchSize : 1000
Upsert : true
UsingLoad : true
IntervalLength : 60
Changed in trafodion: | |
importance: | Undecided → High |
tags: | added: sql-cmp |
Changed in trafodion: | |
assignee: | nobody → Suresh Subbiah (suresh-subbiah) |
milestone: | none → r2.0 |
This problem continues to cause delays in our performance testing schedule -- the latest attempt to create/load the OE_STOCK_512 table resulted in four warehouses missing a group of 364 stock items and one warehouse missing 275 contiguous stock items. This is using the 1.0.1rc3 build (March 3rd, 2015).