Pantheon Community

Preparing for a known traffic spike -- advice or tips?

How to use the Pantheon Platform General Discussion
#1

I’m expecting 700+ concurrent users for a brief 2-3 hour window. I feel like I’ve been getting conflicting info from pantheon: an agency partner rep says that the site will elastically spin up new containers to accommodate traffic spikes; previous interactions with support have told me that I will NOT get any additional php workers and that, therefore, my site will likely choke and die. Can I get a solid answer here? Do I need to crank up my plan or not? It’s a one-time event; most of the year the site receives relatively little traffic.

Some details:

  1. FWIW this is a woocommerce site, and much of the traffic will bypass varnish/the caching layer.

  2. The spike is caused by a first-come-first-serve event ticket release at 5am, which is not a super great idea to begin with, but it’s what it is.

  3. Yes, I have done as much performance tuning and tweaking as is reasonably possible without digging into woocommerce internals or rewriting woocommerce plugins.

  4. Yes, I have been using blazemeter to load-test my test environment with the understanding that the test environment has two php workers, and the live environments on larger plans will have more.

  5. Yes, I have been using new relic to isolate performance bottlenecks.

  6. Right now the site is a performance small.

Any tips or advice? Do I need to upgrade to a larger plan in order to get better performance or not?

#2

The number of application servers your site is given (and other resources, e.g. RAM) is dependent entirely on your service level. Pantheon will not spin up extra servers for you to handle extra load during traffic spikes.

Talk to sales about your ability to temporarily bump your plan up to a higher service level during your high-traffic time period. It is technically feasible to raise and lower your service level; I am unfamiliar with the pricing and packaging of such temporal moves, though.

1 Like
#3

I have not seen any evidence that Pantheon will accommodate service level bumps in any increment less than one month. I would love to be corrected in that!

1 Like
#4

Hi @ianjohn27 – Good question. Yes, because you’re expecting a spike in logged-in users, it sounds like you may want to upgrade your site to a Performance M, L or XL plan, which will increase the number of PHP workers. See https://pantheon.io/pricing-comparison for details on # of PHP workers per plan.

You can do some math to figure out which plan you’ll want. You’ll want to find your site’s response time, which you can find in New Relic.

Here’s the doc on changing plans: https://pantheon.io/docs/site-plan

And a note on doing the math: https://pantheon.io/docs/load-and-performance-testing#calculating-load-capacity-after-launch

The explanation of the math here could use some fleshing out, so let me know if you have further questions.

3 Likes
#5

Hi @ianjohn27, I’m in a similar situation, I’d like to know what experience you had with the spike and what plan you were using. Thanks a lot!

Nicola

2 Likes
#6

Hi Nicola, I cranked up to a performance large plan for the month during the spike. I also disabled every plugin that was not essential to the site’s functioning. I did some basic scripts with loadimpact & loadstorm which allowed me to simulate the flow of a purchaser up to the point of checkout, and then spent a lot of time of New Relic really looking at what was using resources. I found certain plugins were breaking all page caching (yikes!) and which plugins were needlessly consuming resources (some I disabled, and some I just tweaked). I found that switching to a multi-container setup had some issues (I think addressed in the most recent version of woo), but mostly it was pretty smooth.

2 Likes
#7

Is the number of workers per-container, or total for the plan?

Is this thread still accurate?
I’m trying to get a better sense of how the fail-over containers work, for instance
If one of my containers is constrained on PHP workers, does the fail-over start handling requests?
At what point is a container considered failed?