lua_shared_dict mapped to another volume?

gmail · 2018-09-22T00:42:30+00:00

I think i have fixed it and i've created a pull request. https://github.com/openresty/lua-nginx-module/pull/1389On Tuesday, October 16, 2018 at 9:16:23 A...

lua_shared_dict mapped to another volume?

gmail

Hi! I am using Kong which uses OpenResty, and using a lua_shared_dict for cache.

I'm running this on an EC2 instance with 16 GB of included storage and an attached EBS volume that's much bigger (100 GB).

I'd like the lua_shared_dict to be mapped to the attached EBS volume to take advantage of the extra space for cache, but I can't figure out if this is possible, or if it's something transparently handled by OpenResty/NGINX.

It's also entirely conceivable that my mental model of what happens with a lua_shared_dict is completely wrong :) in which case I would love pointers to where I can read more.

Any help gratefully appreciated!

rpaprocki

lua_shared_dicts are memory storage. There are memory zones shared among nginx worker processes. Data is never written to disk; an EBS mount is meaningless. If you want to increase the size of the dictionary you will need more available memory.

Now in theory you _could_ mount a swap partition on an external volume and let the kernel deal with paging that out in order to define a lua_shared_dict larger than the available RAM on the instance... but that probably cracks the top ten of “worst ideas ever”, somewhere between New Coke and wearing a leather jacket on an African safari.

Also, please feel free to head over the Kong Nation (https://discuss.konghq.com/) for kong specific chatter :)

Sent from my iPhone

On Sep 21, 2018, at 09:42, EMM <gm...@alphaorder.com> wrote:

Hi! I am using Kong which uses OpenResty, and using a lua_shared_dict for cache.

I'm running this on an EC2 instance with 16 GB of included storage and an attached EBS volume that's much bigger (100 GB).

I'd like the lua_shared_dict to be mapped to the attached EBS volume to take advantage of the extra space for cache, but I can't figure out if this is possible, or if it's something transparently handled by OpenResty/NGINX.

It's also entirely conceivable that my mental model of what happens with a lua_shared_dict is completely wrong :) in which case I would love pointers to where I can read more.

Any help gratefully appreciated!

.

gmail

Awesome! Usually when I can't find anything on a topic it means my base understanding is Completely Wrong, so thanks for improving my mental model here! This saves me a lot of time (And I think I see more Redis in my future).

Thanks again! (And thanks for the don't-try-this-at-home hack ... sounds like it might even rise to the level of "starting a land war in Asia" :) )

On Friday, September 21, 2018 at 9:50:48 AM UTC-7, rpaprocki wrote:

lua_shared_dicts are memory storage. There are memory zones shared among nginx worker processes. Data is never written to disk; an EBS mount is meaningless. If you want to increase the size of the dictionary you will need more available memory.

Now in theory you _could_ mount a swap partition on an external volume and let the kernel deal with paging that out in order to define a lua_shared_dict larger than the available RAM on the instance... but that probably cracks the top ten of “worst ideas ever”, somewhere between New Coke and wearing a leather jacket on an African safari.

Also, please feel free to head over the Kong Nation (https://discuss.konghq.com/) for kong specific chatter :)

Sent from my iPhone

On Sep 21, 2018, at 09:42, EMM <gm...@alphaorder.com> wrote:

Hi! I am using Kong which uses OpenResty, and using a lua_shared_dict for cache.

I'm running this on an EC2 instance with 16 GB of included storage and an attached EBS volume that's much bigger (100 GB).

I'd like the lua_shared_dict to be mapped to the attached EBS volume to take advantage of the extra space for cache, but I can't figure out if this is possible, or if it's something transparently handled by OpenResty/NGINX.

It's also entirely conceivable that my mental model of what happens with a lua_shared_dict is completely wrong :) in which case I would love pointers to where I can read more.

Any help gratefully appreciated!

.