Openbravo Issue Tracking System - Openbravo ERP
View Issue Details
0039604Openbravo ERPA. Platformpublic2018-11-07 15:032018-11-13 10:47
alostale 
alostale 
immediatemajorrandom
closedfixed 
5
 
3.0PR18Q3.33.0PR18Q3.3 
caristu
Core
Production - Confirmed Stable
2016-08-10
3.0PR16Q4
https://code.openbravo.com/erp/devel/pi/rev/88ea14de1a84e176f457750118a6d4823588d7d2 [^]
No
0039604: can't update information after killing a connection
With Audit trail active, killing DB connections can cause random failures while trying to update information.

Technically the problem is:
1. Pool keeps PooledConnections each of them associated with a physical DB connection
2. When one of those physical DB connections is killed and a PooledConnection is going to be used (borrowed from the pool), Tomcat transparently creates a new physical connection and associates it with the existing PooledConnection instance.
3. Openbravo, in order not to recreate ad_context_info temporary table on each borrow, does it when a new connection is created and marks it as already initialized. This flag is set in the PooledConnection, incorrectly assuming a PooledConnection instance is always tight to the same Connection instance
4. Because of 2nd step, a new uninitialized Connection is associated with an existing PooledConnection marked as initialized, so queries expecting the temporary table to be present will fail.

This situation remains until the connection is finally returned to the pool.
In a Professional instance:
0. Configure DB pool to a fixed 5 connections:
  db.pool.initialSize=5
  db.pool.minIdle=5
  db.pool.maxActive=5
1. As System Admin open Table window
2. Select any table and mark it as fully audited
3. Restart Tomcat
4. Kill the 5 connections that are in pool: in psql execute select pg_terminate_backend(pid), pid from pg_stat_activity where application_name ='openbravo';
5. Log in
6. Open Process Request window
7. Create a new record and save
  -> Saving failed. ERROR: relation "ad_context_info" does not exist Position:
No tags attached.
blocks defect 0039601 closed alostale can't execute DB updates after killing a connection 
Issue History
2018-11-07 15:09alostaleTypedefect => backport
2018-11-07 15:09alostaleTarget Version => 3.0PR18Q3.3
2018-11-13 10:39caristuNote Added: 0107873
2018-11-13 10:46caristuStatusscheduled => resolved
2018-11-13 10:46caristuFixed in Version => 3.0PR18Q3.3
2018-11-13 10:46caristuFixed in SCM revision => a456297ffb38ecfa584a54518d251cc3da9cda12
2018-11-13 10:46caristuResolutionopen => fixed
2018-11-13 10:47caristuFixed in SCM revisiona456297ffb38ecfa584a54518d251cc3da9cda12 => http://code.openbravo.com/erp/backports/3.0PR18Q3.3/rev/a456297ffb38ecfa584a54518d251cc3da9cda12 [^]
2018-11-13 10:47caristuNote Added: 0107875
2018-11-13 10:47caristuStatusresolved => closed

Notes
(0107873)
caristu   
2018-11-13 10:39   
Repository: erp/backports/3.0PR18Q3.3
Changeset: a456297ffb38ecfa584a54518d251cc3da9cda12
Author: Asier Lostalé <asier.lostale <at> openbravo.com>
Date: Thu Nov 08 10:19:51 2018 +0100
URL: http://code.openbravo.com/erp/backports/3.0PR18Q3.3/rev/a456297ffb38ecfa584a54518d251cc3da9cda12 [^] [^]

fixed bug 39603: can't execute DB updates after killing a connection

  Audit Trail infrastructure (ad_context_info in PG) is initialized each time a
  new connection is created. We incorrectly assumed a PooledConnection is always
  associated with a single Connection instance, which is not true in case Connection
  gets corrupted or killed, in this case Tomcat pool transparently re-associates
  a new connection with existent PooledConnection.

  We are now caching in PooledConnection actual Connection to detect this case
  so it can be initialized again.

---
M modules/org.openbravo.apachejdbcconnectionpool/src/org/openbravo/apachejdbcconnectionpool/ConnectionInitializerInterceptor.java
---
(0107875)
caristu   
2018-11-13 10:47   
Code reviewed + tested OK.