Search the Community

Hi. We want to deploy our application on several datacenters, we are going to use the popular multiIP DNS routing(LoadBalancing) approach used for example in Google or large companies. Ie, when a user sends an NS request, the nearest (one of the IP addresses) is returned to him via DNS, thereby the user will get to one of our datacenters (in Lithuania, Finland, Russia, etc.).. In each of the datacenters there are several hyper_server instances with VRRP(keepAlived) – a common IP, if one of the hyper_server instances refuses VIP service, the second instance will be responsible, while it will access the same nodes (slave) the same as the first one. Because we plan to use the Rewrite load balancing mod and assign a unique white address to each working node (slave). I.e. users come with NS: mydomain.com Mydomain.com these are 3 IP addresses (different datacenters): { 1. XXX.XXX.XXX.XXX – Finland(a) 2. XXX.XXX.XXX.XXX – Lithuania(b) 3. XXX.XXX.XXX.XXX – Russia © } DataCenter #1(a) (for example Finland): Common IP master – mydomain.com = XXX.XXX.XXX.XXX – Finland { Master#1 – keepAlived IP – Active Master#2 – keepAlived IP – Reserve Master#3 – keepAlived IP – Reserve } All master servers within the datacenter have the same configuration for working nodes (slave): [server-0], [server-1],[server-2],[server-3]..[server-n] In Finland (slave), the nodes will be as follows: Slave – subdomain.mydomain.com = XXX.XXX.XXX.XXX – Finland { Slave #1 - a0.mydomain.com Slave #2 – a1.mydomain.com Slave #3 – a2.mydomain.com Slave #4 – a3.mydomain.com ... n } (If one of the slave servers falls, it's not as scary as if the master falls and the automatic reserve does not rise.) Then, of course, there is a VRRP for several PgBouncer (for a single pool of DBMS connections), followed by DBMS clusters with sharding and replication. (with auto-failower) All this architecture is not problematic except for one issue: Will we be able to organize a hot backup via master hyper_server via VRRP? Did someone do it? How to automatically raise another master nearby (with the same IP – VIP)(he will immediately pick up all the nodes(slave VM)), does anyone have experience?

Hi All, I would appreciate any help/tips/guidance Bit of a long one, as I have tried to be as detailed as I can - but please stay with it! My setup: Delphi Version - 10.2 Tokyo UniGUI version - 1.0.0.1425 Data Access – UniDAC DBMS – MS SQL Server Deployment – ISAPI dll, IIS 8 (settings exactly match the deployment guide) Number of Users – ~100-120 concurrent (Stress test) Architecture – Compiled to 64-bit - I use one application data module per software module (I have 5), I use CreateOnDemand = True, and Free the data module when the module is closed. I also use EnableSynchronousOperations, and TerminateOnBrowserClose ------------------------------------------------------------------------------------------------------------------------------ Ok, so I have an application that is nearing the end of the first phase of work – which contains all functionality for the agreed minimal viable product. On the surface everything looks great and everything is working as far as I can see. Today I tasked myself with doing some stress tests and looking at memory to make sure everything was stable, before putting the whole app through test. Which leads me to some questions – A single user idles at ~8 mb – once the application is logged on) Upon observing task manager, I noticed the "w3wp.exe" IIS Worker Process remains running in the task list, even after all of the sessions have ended – I guess this is because the deployment guide says to turns off Application pool recycling. The issue I have here is that while all 100 users are connected my RAM is ~300 mb – once they all come off this drops to ~220mb – and the "w3wp.exe" worker process memory usage stays at this indefinitely – and I am stumped to as why this might be the case? I would of thought once all sessions had terminate this would at least drop in RAM usage – I understand the "w3wp.exe" worker process will stay active until recycled, but the deployment guide said to turn off auto/timed/etc recycling – as this could affect the stateful nature of sessions. I get the same affect with using less users, but obviously the RAM usage is much lower, so its less noticeable. I do not get any errors on the stress test, and keep it really simple – log in to the application and open all modules, and scroll through some master detail relationships, to cause the app to do some work. With all this in mind, what is the correct solution to make sure my server doesn’t run out of RAM without constant maintenance and recycling the application pool manually? Am I missing something, or am I doing something wrong? Thanks, Dan