Hanging after Save Server

[mce]

Hi,

I am new in TM1.
I wanted to save data for the TM1 server instance.
I selected the server name in Architect and then Server->Save Data.
Half an hour passed, it is still hanging as if it is doing something, although there is no CPU usage in task manager.
What can I do?

Regards,

mce

[rkaif]

There could be few things behind this.

Most probably, your data in the RAM would be quite large and TM1 is trying to save it in the Server disk. Thats is why it is taking that long.

[mattgoff]

When you do a SaveDataAll, TM1 rewrites all cubes from scratch to the disk (then deletes old .cub and renames the .cub$). Most of the time is taken while the disk subsystem does the writes, not any CPU activity. If you have very large cubes, this is probably the reason for an apparent hang. I'd check perfmon to confirm.

Matt

[lotsaram]

When you do a SaveDataAll, TM1 rewrites all cubes from scratch to the disk

Not quite. During a data save TM1 only re-writes cubes to disk that have data changes, any unchanged cubes are skipped.

The save data places a lock on each and every object on the server which is not released until the entire operation is completed. If you have a lot of data changes and many GB of cubes then the save to disk can take a while. All client sessions will be locked during time and appear to "hang". However the server is only really truly hanging if it encounters lock contention and decides to rollback in which case the save data can take a v very long time to complete. You can check in the server message log file if there are an y rollbacks and if the server is actually truly hanging or whether it is just doing what it needs to do during a data save and behaving as expected.

It would be very good indeed if it was possible to specify a save on specific objects, but unfortunately it is all or nothing.

[Alan Kirk]

I am new in TM1.
I wanted to save data for the TM1 server instance.
I selected the server name in Architect and then Server->Save Data.
Half an hour passed, it is still hanging as if it is doing something, although there is no CPU usage in task manager.
What can I do?

In addition to the replies that you've already received, it's known that servers do sometimes hang when doing a data save. It's why we were never able to use the TI save data command in 8.2.12 (because you could never tell when it was going to happen, though it seemed to be more common with high memory usage) however problems were reported with it as late as 9.1.4. It seemed to be less common when triggering the data save manually, but wasn't unknown. I notice that you're on 9.0, so you may well be encountering this particularly if you're on 32 bit.

You might also want to keep TM1 Top running in the background; it might at least give you some insight as to what (if anything) is happening on the server when something like this happens.

[Martin Erlmoser]

how large is your tm1s.log?

[mce]

Thanks for all the valuable replies.
I found in the forum in another topic that all metadata changes are saved to disk immediately. Only data changes stay in the memmory and saved into the memmory with Server-Save Data process.
Since we did not have data changes that we should not loose, I decided to stop the service for TM1 instance. However it stayed in Stopping status and did not stop. Therefore I restarted the whole server, which fixed the problem as usual.

TM1top could be very useful, but I understand that it is not available in our version of TM1, which is 9.0.

Regards,

[mattgoff]

When you do a SaveDataAll, TM1 rewrites all cubes from scratch to the disk

Not quite. During a data save TM1 only re-writes cubes to disk that have data changes, any unchanged cubes are skipped.

Fair correction. In our case this is usually most of our cubes, and definitely all of the big ones. Note that it doesn't matter how many changes you make, if you change even one cell, the cube is completely rewritten to disk. Reminds me of the old days of MS Office. "Save" did a full file rewrite unless you ticked the "Fast Save" checkbox. "Fast Save" would just append a diff to the end of the file, trading file space for speed. Floppies were sloooooow!

Speaking of rollback, I never saw when/if this was fixed, but the last I heard from Cognos was that SaveDataAll should be in a process by itself in a chore by itself. Doing otherwise can cause rollback loops.... I think this issue was introduced in 9.1 with the locking model, so the OP should be fine, but something to consider for people on 9.1+.

MCE: It's in Stopping state for a long time because the first thing the service does at shutdown is do a SaveDataAll. There's no way around it without force-killing the process. TM1Top is available for 9.0 but you have to get it from support directly. It wasn't bundled with the installer until later.

Matt

[Alan Kirk]

TM1top could be very useful, but I understand that it is not available in our version of TM1, which is 9.0.

TM1 Top was introduced, from memory, in 8.4 but you're correct that it isn't part of the standard install in 9.0. However rather than chasing around IBM support trying to find the vanilla one it would be faster to download Ben Hills' enhanced version of it from the thread http://forums.olapforums.com/viewtopic.php?f=21&t=553.

I'm using it with a 9.0 installation as we speak.

[Alan Kirk]

It's in Stopping state for a long time because the first thing the service does at shutdown is do a SaveDataAll. There's no way around it without force-killing the process.

The other thing is that if you have more than a trivial amount of data to save, the service will report "Oh, no, no response!" and throw its little girly-man hands in the air within seconds of you telling it to stop. What I usually do is to do a data save before I stop the service so that there's nothing to save. Even then you sometimes get a dialog that the service is "not responding in a timely fashion" but it makes no difference because you can see the memory use falling off a cliff and know that the server is safely down. I just click on the resulting dialog to ignore it.