Fork me on GitHub
#portkey
<
2018-03-23
>
viesti14:03:07

> Each Execution Context provides 500MB of additional disk space in the /tmp directory. The directory content remains when the Execution Context is frozen, providing transient cache that can be used for multiple invocations.

viesti14:03:53

Thinking that when we get to implementing delta redeploy (https://github.com/portkey-cloud/portkey/issues/10) we can cache precompiled classes in /tmp

viesti14:03:07

and fetch only newly compiled classes

viesti14:03:55

hmm, although /tmp is probably ec2 dependend

viesti14:03:18

if a new invocation ends up creating a process on a separate machine, cache won’t be there

viesti14:03:51

another thing:

viesti14:03:52

> Background processes or callbacks initiated by your Lambda function that did not complete when the function ended resume if AWS Lambda chooses to reuse the Execution Context. You should make sure any background processes or callbacks (in case of Node.js) in your code are complete before the code exits.

viesti14:03:09

this might mean that this could be done on the background…

viesti14:03:47

wonder how billing goes in that situation…

viesti14:03:54

hum, I think we/you (don’t remember author of the idea) had such background loading idea

viesti14:03:03

even inside the lambda process

cgrand14:03:07

On bg process, I think it’s clear: when the main processing halts they are frozen

viesti14:03:17

hmm, might be so

viesti14:03:35

resume if AWS Lambda chooses to reuse the Execution Context.

cgrand14:03:40

(a week or so ago I researched bout them)

viesti14:03:42

stated by that

cgrand14:03:28

so basically if you have a heavy bg process it’s going to slow down your handler and thus you’ll be billed for the whole combined esexcution time that’s all

viesti14:03:12

apropo, was working on reading models from aws-java-sdk-models for aws-clj-sdk, but got bogged down on some new api, thinking about just skipping those that need changes to be supported

cgrand14:03:33

delta redeploys are two fold: • upload less • do a “soft” redeploy

viesti14:03:21

soft would be for 5min, but thinking that if same compute instance is reused, jvm might be warm too

viesti15:03:33

how's the idea on instrumenting java.security factories?

cgrand15:03:02

too busy to implement it

viesti15:03:51

if I'd be able to infuse the weights of your neural net to be able to implement/help in implementing :)

cgrand15:03:19

we don’t have an issue open for these cryptos?

viesti15:03:47

aah, that I can do

viesti15:03:06

with less training on my net

viesti15:03:06

there's a test though, I'll link to it

viesti15:03:53

had in mind to do a bot that watches CircleCI build status (since can't add CircleCI integration to Slack)