Skip to main content
Question

Has the concept of "transaction" changed since UV 11.3.x

  • Oct 27, 2026 3:29 AM UTC
  • 4 replies
  • 115 views

Gregor Scott
Forum|alt.badge.img+4

We are attempting the adoption of UV 14.2.1 after being stable with 11.3.5 for a while.

It is not going very well. The change between UV 11 and UV 14 is significant and I am not able to find much supporting material to help guide our adoption.

After reviewing the available docs and following the guidance for tuning the `uvconfig` settings we have our target base system configuration established.

Connecting a single uvsh to run `UPDATE.ACCOUNT` results in "The GLM lock table is full." messages.

The guidance for some of the `uvconfig` settings is to reference the maximum number of locks in a transaction.

Seems simple enough, if taking the UV 11.3.5 perspective of a transaction being instigated by BASIC code using the transaction semantics. That is how we approached the setting determination and the results indicate this is not correct.

When digging into some of the UV 14 GLM related changes there seems to be another concept for a transaction.

Using the uvglm_tool utility it is possible to list active transactions on a system.

This is the output, on a system with 0 client connections:

 

# uvglm_tool

What information to display?

 0: quit

 1: Show the Map of Locks

 2: Holding Locks and Waiting Queues for a Given Transaction

 3: Holding Transactions and Waiting Queues for a Given Lock

 4: More Options for Developer

selection = 4

displaying GLM structure by attaching to shared memory.

What information to display?

 0: Back to previous menu

 1: transaction table

 2: detailed info about one transaction

 3: detailed info about all transactions

 4: all locks of a (normal) file by one transaction

 5: (normal) resource table

 6: detailed info about one (normal) resource

 7: detailed info about all (normal) resources

 8: waiting transactions

 9: table info (latches, lock counts, free nodes)

selection = 9

Which table to display?

0: Back to previous menu

1: Transaction Table

2: Transaction Table Summary

3: Lockname Table

4: Lockname Table  Summary

selection = 2

Transaction Table Summary:

-------------------------------------------------------

num of transaction entries                  =    256

num of active transactions                  =    2

avg num of locks per transaction            =    0.000000

max num of locks by a transaction           =    0

min num of locks by a transaction           =    0

avg num of free own_wait nodes              =    0.000000

Which table to display?

0: Back to previous menu

1: Transaction Table

2: Transaction Table Summary

3: Lockname Table

4: Lockname Table  Summary

selection =
 

How can there be 2 active transactions without any active clients?

So I am now wondering if the guidance on tuning some of the uvconfig settings is based on a different concept of a transaction. Can anyone shed light on this?

 

4 replies

Forum|alt.badge.img

Rocket Support will review the uvdiag and other results as part of the support case before advising further.  Next step is to reproduce the behavior with UPDATE.ACCOUNT internally first.


Forum|alt.badge.img

See the support case for more details.  ‘transaction’ in this situation references to GLM lock bucket transactions and not Basic transactions.  The ‘GLM lock table’ full message happens in this situation because GLM_MEM_ALLOC is set extremely high.  The recommendations are to keep N_GLM_SELF_BUCKET and GLM_MEM_ALLOC at the default sizes, then change ‘N_GLM_GLOBAL_BUCKET to the prime number nearest to NUSERS * 3’ (per the documentation).

Based on the reproduced UPDATE.ACCOUNT scenario, each time a GLM global bucket is populated either initially or when additional nodes are needed, the number of GLM_MEM_ALLOC allocations are done.  If we assume an average of 60 bytes per GLM_MEM_ALLOC allocation, the default 160 MB lock table size is exceeded (16 * GLM_MEM_SEGSZ).  In formula form: (GLM_MEM_ALLOC * N_GLM_GLOBAL_BUCKET * 60) was greater than (16 * GLM_MEM_SEGSZ), hence the error.  The noted 60 bytes per allocation, is only an estimate with testing in this one scenario, due to other lock table conditions this number may be different.   


Gregor Scott
Forum|alt.badge.img+4
  • Author
  • Participating Frequently
  • Oct 28, 2026 2:02 AM UTC

Thanks for the update ​@Brian Kupzyk 

I thought there must be another “transaction” concept being referened.

We are coming directly to UniVerse 14.2.1 from 11.3.5, and my understanding of a “transaction” was BASIC code related. That obviously does not apply to uvconfig setup determination under 14.2.1.

 

The default uvconfig settings for these parameters are:

# N_GLM_SELF_BUCKET - The number of hash buckets for the per-process locking

# table. This parameter is highly application dependent,

# If the application requires a large number of locks in

# one transaction (more than 20), you should increase this

# setting to the closest prime number to the maximum number

# of locks per transaction.

N_GLM_SELF_BUCKET 23

# GLM_MEM_ALLOC - The number of lock nodes allocated for each memory

# request. This parameter is highly application dependent.

# if the application requires a large number of locks in

# one transaction, this setting should be increased to the

# maximum number of locks per transaction * 2.

GLM_MEM_ALLOC 10

 

The description in the “What_You_Need_to_Know_When_Moving_from_UniVerse_11_to_14.pdf” for these settings is:

o N_GLM_SELF_BUCKET. Sets the number of hash buckets in the per-user 
locking table. Adjust this based on the number of concurrent locks a user 
may hold.
o GLM_MEM_ALLOC. Specifies the number of lock nodes available to a uvsh 
session. Increase this value if sessions hold many locks concurrently. A good 
rule of thumb is to set it to twice the maximum number of locks held per 
transaction or session.

 

The document “universe_administering_universe_on_windows_and_unix_platforms_version_14.2.1” gives the following guidance for these settings:

GLM_MEM_ALLOC 
The number of lock nodes allocated for each memory request. This 
parameter is highly application dependent. If the application requires a large 
number of locks in one transaction, this setting should be increased to the 
maximum number of locks per transaction * 2. The default is 10. 

N_GLM_SELF_BUCKET 
The number of hash buckets for the per-process locking table. This 
parameter is highly application dependent, If the application requires a large 
number of locks in one transaction (more than 20), you should increase this 
setting to the closest prime number to the maximum number of locks per 
transaction
. The default is 23. 

 

So everywhere I look I am advised to adjust these settings based on locks per transaction or session. There is no advice on how to measure the the maximum locks within this new “transaction” concept.

As there is now an additional “transaction” concept I need to cater for, if I adopt your recommended approach of just using the defaults then how do I measure the “maximum locks per transaction” for this new GLM-related transaction to identify if/when these default value should change?

Or is the advice in the above sources out of date and these settings just should not be changed?

 


Forum|alt.badge.img

The wording in all of the sources related to these GLM and AFT settings could be improved. I will work on it as time permits.

A GLM "transaction" is a user or phantom session. Each owner (GLM transaction) has one entry in the per-process lock table, so "locks per transaction" means "locks currently held by one session". N_GLM_SELF_BUCKET and GLM_MEM_ALLOC do not limit how many locks a session can hold, and the defaults will not cause lock failures. A very high GLM_MEM_ALLOC, as in the earlier case, can use up GLM shared memory and cause the "GLM lock table is full" error.

Both tables below are snapshots. Take each one several times while your heaviest job is running.

Locks per session: run uvglm_tool, choose 4 (More Options for Developer), then 9 (table info), then 2 (Transaction Table Summary):

Transaction Table Summary:
-------------------------------------------------------
num of transaction entries                  =    256
num of active transactions                  =    2
avg num of locks per transaction            =    1.000000
max num of locks by a transaction           =    2
min num of locks by a transaction           =    0
avg num of free own_wait nodes              =    0.031250

The "num of transaction entries" value matches NUSERS. Keep the largest "max num of locks by a transaction" value. Raise N_GLM_SELF_BUCKET only if sessions regularly hold well over 20 locks and you care about lock lookup speed. Use a prime near the measured maximum.

Memory held in bucket freelists: run uvglm_tool, choose 4, then 9, then 4 (Lockname Table Summary). For each node size it shows the average, maximum and minimum free nodes per bucket. Ignore any size whose maximum stays at 0 across all snapshots, because that size is not used from the bucket freelists. For the others, compare the maximum with your GLM_MEM_ALLOC value. At the default of 10 you will normally see about 9 or 10. If you have raised it and the maximum is close to your value, a bucket has taken a full batch and is holding the unused nodes. A size that was non-zero earlier but drops to 0 during the run suggests that freelist is being drained and refilled.

Leave GLM_MEM_ALLOC at 10 unless profiling shows contention from frequent refills. If you change it, increase it modestly, and I would avoid the "twice the maximum locks" suggestion for now. If you have raised it and the freelists hold far more than your workload needs, lower it, but not below 10.