Welcome to the new Gigaspaces XAP forum. To recover your account, please follow these instructions.

Ask Your Question
0

topology question

Hi All, I'm new to GS. I have topology/architecture question: I have two transient spaces in cluster over WAN (asynch replication). When no connection(network failure) between them for some time(sometimes day) - redo log become very large and it influence badly on perfomance. If I lmit redo log , after network is established - spaces not synchronized. How I can reslove this problem : redo log will not cause perfomance problem (not grow) and after reconnection two spaces would be synchronized ?

Thanks, Michael

Edited by: miki vin on Apr 6, 2008 11:25 AM

{quote}This thread was imported from the previous forum. For your reference, the original is [available here|http://forum.openspaces.org/thread.jspa?threadID=1924]{quote}

asked 2008-04-06 10:18:56 -0500

mikivin gravatar image

updated 2013-08-08 09:52:00 -0500

jaissefsfex gravatar image
edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
0

Miki,
This is a known issue with long disconnection scenario.
How do you limit the redo log size? Are you using the redo-log-capacity cluster parameter?
Why you have such long disconnections?
Can you restart one of the spaces to sync with each other ? This will invoke full recovery process.
Another option to resolve this is to move into persistent space mode. This will impact the performance.
We have made few important fixes lately with such config. Contact support for exact build to use in such a case. Another simpler way to resolve this is to have larger JVM (64 bit).
Last option is to limit the space size using relevant cache size or memory usage settings. In this case , adding new entries to the space will be blocked and the space will throw MemoryShortageException once you will breach the upper threshold.

Please note the redo log is used also after a space have fully recovered. In this case the redo log is cleared before the memory recovery started and it is filled with the in-flight events happening during the memory recovery phase (full snapshot replication of the primary to the replica). Since the memory recovery phase is relatively short (and made even shorter with 6.5 using parallel recovery mechanism) , the redo log should have a size that would accommodate enough entries that would be accumulate during the memory recovery phase.
For example : if the memory recovery takes 30 seconds, and there are 200 new entries written to the space every second , the redo log size should be at least 6000.

We will enhance the reconnection mechanism in future versions to disable to accumulation of destructive operations into the redo log after some time (configured) in order it won't be filled up. In case a case a full memory recovery will be initiated once the connection will be re-established. Another option would be to persist only the redo log data.

Shay

answered 2008-04-07 13:10:48 -0500

shay hassidim gravatar image
edit flag offensive delete link more

Comments

Shay, Thank You a lot for answering. 1.I use 'redo-log-capacity' in cluster parameter . Problem with it - after log is full spaces will not be synchronized correctly. 2.Any solution that cause spaces will be usynchronized or data lost - not suitable for system.(space size,restart spaces) 3.We can't move to persistent space mode - performance.

I would ask if Mirror service can help in this situation ? Can this problem be resolved by any architecture? Multi-Clustrer using mirror ?

Thanks aagin Michael

mikivin gravatar imagemikivin ( 2008-04-08 05:33:53 -0500 )edit

If your system got such long disconnections you should construct it using 2 different clusters and add a mechanism that will sync each site when ever the connection is valid.

See: http://www.gigaspaces.com/wiki/displa... the "Multi-Cluster Architecture over WAN using IWorker" and "Multi-Cluster Architecture Over the WAN using Mirror Service" might be relevant in your case.

Shay

shay hassidim gravatar imageshay hassidim ( 2008-04-10 05:56:23 -0500 )edit

Thank You very much.
Michael

mikivin gravatar imagemikivin ( 2008-04-14 06:49:43 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

1 follower

Stats

Asked: 2008-04-06 10:18:56 -0500

Seen: 32 times

Last updated: Apr 07 '08