October/November 2013
October
- Heavy-load DML test on database ("IDX_UNDER_LOAD") with 10^6 records to check whether there are bugs when all connections performed via FB3 client (450 connections)
Result: deadlock with MON$ tables detected, fixed by Vlad Khorsun
- Repeat heavy DML-test without querying MON$-tables, 450 connects, all via FB3 client
Result: no problems
- Apply patches to avoid deadlocks while querying MON$ tables. Heavy DML (350, 450, 550 connections) + 4 separate scripts with querying MON$-tables + connection time calibration script
Result: huge amount of too long connections (over 8 seconds), several server crashes
- Repeat heavy DML-test with v2.5 clients only
Result: server does not respond after ~30 minutes of runtime
- Several attempts to get a coredump when the server crashes under high load (~100 connections)
Result: race condition detected, fixed by Alex Peshkoff
- Repeat heavy DML-test with fresh patches, mixed v2.5 and v3 clients (450 connections)
Result: too long connections (over 60 seconds) detected
November
- Bugcheck 165 (cannot find tip page) happens if engine crashed right after CREATE DATABASE
Result: CORE-4260, fixed by Vlad Khorsun
- Partially ported (schema and a number of queries) advanced SQL language exercises to Firebird, intended to test both old and new query data access methods, results are compared with Oracle and MSSQL
Result: a number of issues detected, see below
- Found test case delivering wrong result using a left join
Result: CORE-4261, fixed by Dmitry Yemanov
- Found test case raising error: "context already in use"
Result: CORE-4262, fixed by Dmitry Yemanov
- Checked that CORE-4236 (Database shutdown is reported as successfully completed before all active connections are in fact interrupted) is really fixed
Result: confirmed
- Investigated strange cases of 'lock conflict' when script contained only inserts (and later no DML at all)
Result: CORE-4265, fixed by Dmitry Yemanov
- Found test case causing wrong output when result of windowed function is used in query with useless WHERE 0=0 and GROUP BY clause
Result: CORE-4269
- Found test case causing wrong output (no rows) in case of subquery with windowed function + where IN (select ...)
Result: CORE-4270, fixed by Dmitry Yemanov
- Profiled the engine trying to find the reason of high connections times under load
Result: found a few problematic places, reported to the core devs, proved that only SS (shared cache arch) is affected
Pavel Zotov
Moscow, Russian Federation
October/November 2013
October
- Heavy-load DML test on database ("IDX_UNDER_LOAD") with 10^6 records to check whether there are bugs when all connections performed via FB3 client (450 connections)
Result: deadlock with MON$ tables detected, fixed by Vlad Khorsun
- Repeat heavy DML-test without querying MON$-tables, 450 connects, all via FB3 client
Result: no problems
- Apply patches to avoid deadlocks while querying MON$ tables. Heavy DML (350, 450, 550 connections) + 4 separate scripts with querying MON$-tables + connection time calibration script
Result: huge amount of too long connections (over 8 seconds), several server crashes
- Repeat heavy DML-test with v2.5 clients only
Result: server does not respond after ~30 minutes of runtime
- Several attempts to get a coredump when the server crashes under high load (~100 connections)
Result: race condition detected, fixed by Alex Peshkoff
- Repeat heavy DML-test with fresh patches, mixed v2.5 and v3 clients (450 connections)
Result: too long connections (over 60 seconds) detected
November
- Bugcheck 165 (cannot find tip page) happens if engine crashed right after CREATE DATABASE
Result: CORE-4260, fixed by Vlad Khorsun
- Partially ported (schema and a number of queries) advanced SQL language exercises to Firebird, intended to test both old and new query data access methods, results are compared with Oracle and MSSQL
Result: a number of issues detected, see below
- Found test case delivering wrong result using a left join
Result: CORE-4261, fixed by Dmitry Yemanov
- Found test case raising error: "context already in use"
Result: CORE-4262, fixed by Dmitry Yemanov
- Checked that CORE-4236 (Database shutdown is reported as successfully completed before all active connections are in fact interrupted) is really fixed
Result: confirmed
- Investigated strange cases of 'lock conflict' when script contained only inserts (and later no DML at all)
Result: CORE-4265, fixed by Dmitry Yemanov
- Found test case causing wrong output when result of windowed function is used in query with useless WHERE 0=0 and GROUP BY clause
Result: CORE-4269
- Found test case causing wrong output (no rows) in case of subquery with windowed function + where IN (select ...)
Result: CORE-4270, fixed by Dmitry Yemanov
- Profiled the engine trying to find the reason of high connections times under load
Result: found a few problematic places, reported to the core devs, proved that only SS (shared cache arch) is affected
Pavel Zotov
Moscow, Russian Federation