Thanks ,
Stephen
Based on your suggestion, we are optimizing for the first and second issues. 
For the third suggestion, our business has many 'temporary tables',  which have 
different column and be detroyed when used. So I'm afraid it is hard for us to 
avoid create/destory tables.
The column is dynamic so it cannot be written  in XML either . I’ not sure  
whether  createCaches()  can meets the requirements or not . 

 

Thank you for your help!



At 2023-10-17 17:18:56, "Stephen Darlington" <[email protected]> wrote:

There's a lot going on there, and I don't have the time to fully analyse, but I 
will note a few things:
You have a lot of very poorly optimised queries (full table scans)

I see a few "long JVM pauses" which suggests poorly configured garbage collector
Ignite is not designed for large numbers of cache creation/destruction. Lots of 
simultaneous queries are fine, because those do not require a global lock

If you need to create a lot of caches at once, you can use XML or a programming 
language and use the createCaches() method





On Tue, 17 Oct 2023 at 10:43, y <[email protected]> wrote:





Here is cluster log(from one node) and createtable sql(Appendix).
The Cluster has 10 nodes and do stress testing with 50 thread .It means there 
50 query or createtable operations at the same time. Notice that when create 
table,the cluster node log has many ‘Thread - WAITING’. That's not normal, 
right?







At 2023-10-17 15:19:33, "Stephen Darlington" <[email protected]> wrote:

Can you share some more information about your cluster? There is no way that 
creating a cache should take so long.


On Tue, 17 Oct 2023 at 03:51, y <[email protected]> wrote:

Hello.
Everyone!

  Creating table statements is executed synchronously and will block other DDL 
statements.  There are serious performance issues in concurrent environments. 
It takes one minute to create one table. 
   If there is any way to solve this problem, such as changing it to serial 
execution?



Thanks,

Tianyu-Hu



  

Reply via email to