A customer had a particularly difficult and awkward problem. A number of user reported that they could not see the status updates of others users who were part of the same network.
This seemed to affect a handful of users in the following pattern. User A cannot see User B’s status updates whilst User C can see User B’s updates. The problem is associated with User B. To c0mplicate things further User A shares a network with User B but does not follow him whilst User C also shares a network but does follow User B. So if you follow User B you get to see their status updates in your Status Updates. You should not need to follow the user to see their status updates the fact that you’re sharing a network should be sufficient.
I logged a PMR which thankfully found it’s way into the ever helpful hands of David McCarthy. An ISSC colleague mentioned that this had been seen in IBM’s deployment of Connections (w3) by Luiz Benietz and others so getting to the bottom of this would kill two birds with one stone.
IBM had me extract data from PEOPLEDB and HOMEPAGE databases using the DB2 commands below to provide the PERSON_ID of the users within HOMEPAGE to make sure it matches with the PROF_GUID within PEOPLEDB which for User B is E833339D-AED2-425F-8600-64CEFD85A3A5.
Comparison of NR_NETWORK showed that the data in there was correct.
PROF_GUID = E833339D-AED2-425F-8600-64CEFD85A3A5
PERSON_ID = 0bdd7880-8d9a-4fe1-ab13-77c54bb429bc
EXID = E833339D-AED2-425F-8600-64CEFD85A3A5
Using the PERSON_ID I was then able to retrieve data from the HOMEPAGE.NR_NEWS_STATUS_NETWORK table which details all the status updates posted by that PERSON_ID which in the HOMEPAGE.NR_NEWS_STATUS_NETWORK table is identified by ACTOR_UUID. This table also shows the READER_ID (detailed below) as well as a BRIEF_DESC column which shows the first 200 (or so) characters of the status update.
db2 “select * from HOMEPAGE.NR_NEWS_STATUS_NETWORK where ACTOR_UUID = ‘0bdd7880-8d9a-4fe1-ab13-77c54bb429bc‘”
ACTOR_UUID = 0bdd7880-8d9a-4fe1-ab13-77c54bb429bc
READER_ID = PERSON_ID of the person that is consuming the status updates ie those listed in the READER_ID column should be able to see the status updates.
To find out the names of the people who can read the updates you will need to use the READER_ID from HOMEPAGE.NR_NEWS_STATUS_NETWORK and find out the name of the user from the HOMEPAGE.PERSON by using the READER_ID converted to PERSON_ID.
db2 “select DISPLAYNAME from HOMEPAGE.PERSON where PERSON_ID = ‘714fbcd9-1740-477b-972d-a6456a35e4ea'”
I contacted the user and they could indeed see User B’s status updates.
To see how many people User B has in their network the following command was run which produced a value of about 140.
db2 “select count(*) from HOMEPAGE.NR_NETWORK where PERSON_ID = ‘0bdd7880-8d9a-4fe1-ab13-77c54bb429bc‘”
IBM asked for the following trace to be enabled whilst User B posted a status update and in the trace.log an exception was seen during the processing of one of the threads.
*=info: java.sql.*=all: com.ibatis.sqlmap.*=all: com.ibm.lconn.homepage.*=all: com.ibm.lconn.news.*=all
It was suggested that 126.96.36.199 be installed but that would not be possible. The following fixes were applied which aimed to improve that code path to handle the case where the use records in the Homepage database are out of sync.
188.8.131.52 LO63965 VBUT8LNH89 Enhancement to News aggregation service for better handling of exceptional cases where user records become inconsistent with Profiles
184.108.40.206 LO66468 SCRD8QCDDK Profiles Sync issue. Too many requests made on Event consumption to Profiles colleague feed
Unfortunately the problem was still apparent after the iFixes so more debugging was enabled, this time com.ibm.lconn.news.*=all: com.ibm.lconn.events.*=all
Thankfully this time the trace.log provided much more valuable information.
[30/05/12 16:14:59:301 BST] 00000067 StatusUpdateS 3 com.ibm.lconn.news.service.impl.StatusUpdateService insertStatusUpdate Going to insert into db with a batch process this number of records: 144 and BUCKET_SIZE_BATCH_INSERT: 100
[ … snip …]
[30/05/12 16:14:59:308 BST] 00000067 NewsStatusNet 3 com.ibm.lconn.news.data.dao.impl.ibatis.NewsStatusNetworkDao insertBatch – doInSqlMapClient rowsaffected 100
[30/05/12 16:14:59:308 BST] 00000067 NewsStatusNet < com.ibm.lconn.news.data.dao.impl.ibatis.NewsStatusNetworkDao insertBatch RETURN 100
[30/05/12 16:14:59:312 BST] 00000067 StatusUpdateS < com.ibm.lconn.news.service.impl.StatusUpdateService insertBatch RETURN 100
[ Here – it’s missing the insert for the remaining 44 records ]
[30/05/12 16:14:59:312 BST] 00000067 StatusUpdateS 3 com.ibm.lconn.news.service.impl.StatusUpdateService insertStatusUpdate Going to insert into db with a batch process this number of records: 3 and BUCKET_SIZE_BATCH_INSERT: 100″
The response from L3 was:
The root cause is a coding error in a routine performing batch insert. The issue should be reproducible for people in the following conditions:
- The number of people in the intersection of the sets (colleagues, followers) for the author of the status update is > 100
- The number of colleagues – number of people in the intersection is > 100 (this is what happen for this very specific PMR)
- The number of followers – number of people in the intersection is > 100.
So the problem we are seeing is because the user had more than 100 people in their network and/or followers and the code would not add the 44 others to the database. pressing further I asked for an iFix for this which is too complicated and the only option is to go to 220.127.116.11 CR1.
“Upon further review – the issue appears to be indirectly fixed in 18.104.22.168….We performed a code review in 22.214.171.124 – this code path was entirely re-written. The final section of the batch is inserted in the new code in 126.96.36.199.”
Up until this point the problem still hasn’t been resolved as updating Connections has not been planned yet but IBM support are fairly confident that 188.8.131.52 CR1 will iron out this. With regards to w3, David McCarthy was able to track down one user who reported the problem initially who now (after the upgrade) can see all status updates.