New hardware has been purchased for the system that hosts the LUV VM. The old server is a i7-920, 8G of RAM, and 2*750G SATA disks. The "new" server is a i7-930, 48G of RAM, 2*250G SATA SSD, and 2*2TB SATA disks. I put new in quotes because it's not new hardware, it's hardware someone else rented first. For most hosting providers getting hardware someone else used first is standard practice. But the Hetzner business model is that the default is to buy new systems that haven't been used before. Systems that have been used are sold by a reverse-auction system where the price on each system goes down steadily until someone buys it. I am not sure exactly when I will migrate the LUV VM, but it will probably be late at night. I will coordinate with the committee because there will be a time when we have 2 separate instances of Drupal and I don't want anyone to make changes to the old one that get lost. The advantage of doing it this way is that for the people viewing the web site there will be no outage. Before starting the changes I will stop Postfix (which will only be noticed as delayed list mail) and configure Apache to fail any requests to the list server configuration web pages. The main aim is to have the minimum of surprises for people who don't read my announcement messages. Delayed mail is not noteworthy. A list server web page being down for a while usually isn't a big deal. The main web site with information on future meetings etc is the important thing that needs to remain up. The current LUV VM has 2G of RAM and 512M of swap of which 450M is used. The new VM will have at least 4G and maybe 6G of RAM. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
time when we have 2 separate instances of Drupal and I don't want anyone to make changes to the old one that get lost.
Re the backend MySQL storage, you make the new MySQL instance a slave of the old one. That way nothing can get lost. We've find this many times with large systems. No outage and no data loss. By the way, things tend to actually work out better in many respects to NOT have frontend and database on the same VM. This is generally the first step we take in creating resilient and scalable environments. There are many disadvantages to combining all services onto the same box. In vm hire terms, it may also be cheaper to use more smaller instances as the pricing model favours those. Regards, Arjen.
On Monday, 19 June 2017 8:30:46 AM AEST Arjen Lentz via luv-main wrote:
time when we have 2 separate instances of Drupal and I don't want anyone to make changes to the old one that get lost.
Re the backend MySQL storage, you make the new MySQL instance a slave of the old one. That way nothing can get lost. We've find this many times with large systems. No outage and no data loss.
Thanks for the suggestion, it's something to consider when migrating a system that has a non-trivial amount of writes. But when dealing with a system that has an average of less than 1 write of actual data per week it's not needed.
By the way, things tend to actually work out better in many respects to NOT have frontend and database on the same VM. This is generally the first step we take in creating resilient and scalable environments.
Well if you want some sort of HA scheme that may be the case. But if your service is small enough that the amount of effort required to make a HA system more reliable than a single system isn't viable then a single VM.
There are many disadvantages to combining all services onto the same box. In vm hire terms, it may also be cheaper to use more smaller instances as the pricing model favours those.
https://www.linode.com/pricing For Linode the pricing scales fairly linearly with size for most common sizes and other hosting companies have similar schemes. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
participants (2)
-
Arjen Lentz -
Russell Coker