Reading value of variable fails after some number of requests about spark-server

[wdimmit]

The console log for the server appears as follows at the point of failure. The upper request worked, the lower request did not.

::ffff:67.161.118.40 - - [03/Jan/2017:20:23:37 +0000] "GET /v1/devices/310023000647343432313031/randomSmall?access_token=23ebff850ac11f2ebf20d8683a04a6934ad3f289 HTTP/1.1" 200 13 "-" "curl/7.47.0" GetVar { deviceID: '310023000647343432313031' } ::ffff:67.161.118.40 - - [03/Jan/2017:20:23:38 +0000] "GET /v1/devices/310023000647343432313031/randomSmall?access_token=23ebff850ac11f2ebf20d8683a04a6934ad3f289 HTTP/1.1" 400 58 "-" "curl/7.47.0" GetVar { deviceID: '310023000647343432313031' }

[jlkalberer]

@wdimmit - thanks, we'll look into this. When I moved the code over I only tested functions and not variables. We should get a fix out by Thursday.

[wdimmit]

@jlkalberer For what it's worth, this error state seems to block access to functions as variables once it is triggered.

[wdimmit]

One last detail: This happens with function calls as well. Error message is the same. Log snippet of the break between the last good request and the first bad request follows.

`FunCall { deviceID: '1c0036001947343433313339' }

TODO: Type buildArguments

Pinged, replying { deviceID: '1c0036001947343433313339' }

::ffff:67.161.118.40 - - [03/Jan/2017:20:51:47 +0000] "POST /v1/devices/1c0036001947343433313339/arioSet HTTP/1.1" 200 285 "-" "curl/7.47.0"

FunCall { deviceID: '1c0036001947343433313339' }

TODO: Type buildArguments

::ffff:67.161.118.40 - - [03/Jan/2017:20:51:48 +0000] "POST /v1/devices/1c0036001947343433313339/arioSet HTTP/1.1" 400 58 "-" "curl/7.47.0"`

[jlkalberer]

@AntonPuko - I think this might be the culprit. https://github.com/Brewskey/spark-protocol/blob/master/src/clients/Device.js#L625-L635

I don't think it should reject there. @wdimmit you can validate by turning on showVerboseDeviceLogs in spark-protocol.

[wdimmit]

Setting that flag to true in /src/settings.js had no impact on the console log output. Also tried setting "verboseProtocol: true", but that had no impact either.

[jlkalberer]

@wdimmit - Is this your test firmware? Do you mind making a gist for @AntonPuko to test with?

[jlkalberer]

Also, are you doing multiple requests for multiple devices or just a single device?

[wdimmit]

I'm doing 1 request / second for a single device, with no other activity against the API. Gist here: https://gist.github.com/wdimmit/0b9a86f5f0ccf212f22fdc51601c3e45

Thanks.

[jlkalberer]

Ok, one more thing to check. Have you manually removed the spark-protocol from your spark-server npm modules and done another install? NPM isn't smart enough to know that a git repo has been updated :/

[AntonPuko]

hm..can't reproduce anyof your issues:
requests work fine:

and string vars, too:

[wdimmit]

I didn't realize that a copy was being cached in node_modules.... My mistake. This voids my observation about those settings variables not changing anything.

This fixes the string variable issue. I'm re-investigating the failure after a number of requests issue.

[jlkalberer]

@wdimmit - yeah, that's why on our dev machines we are using npm link and running watchman on the spark-protocol folder. It makes testing a whole lot easier.

[wdimmit]

I'm still seeing the read failure. It often takes several minutes to occur, so 50 iterations will not be sure to trigger it. My last test took roughly 5 minutes to start failing.

I've enabled both the verbosity settings options in spark-protocol (specifically in ./node_modules/spark-protocol/src/settings.js) and I'm still not seeing anything extra in the console surrounding the failure.

[AntonPuko]

Well, I reproduced your requests issue. seems it something about message's token counter, which at some point becomes bigger than allowed in a device. I'll dig into it tomorrow.

[wdimmit]

@AntonPuko Glad you've seen it. If it helps, the magic number seems to be 250 +/- 5. Timing of the requests doesn't seem to matter. Thanks.

[AntonPuko]

ye, noticed that also. Seems the counter breaks when its more than one byte

[jlkalberer]

Ahh, that makes sense. It can be fixed here

For tokens it must be less than 256
https://github.com/straccio/spark-protocol/blob/master/clients/SparkCore.js#L547-L570

Here is what it's supposed to be:
https://github.com/straccio/spark-protocol/blob/master/settings.js#L35-L44

This happened because of the large refactor / simplification :/

[jlkalberer]

Alright, fix is out. I haven't tested as I don't have a device set up but I'm pretty sure we should be good to go after this: Brewskey/spark-protocol@4afd37f

[wdimmit]

It appears the previous commit to spark-protocol (pull request 19....) broke something related to variable reads. All queries return the following, and the pull request in question impacted this area of the code:

{
"error": "device.onApiMessage is not a function",
"ok": false
}

[AntonPuko]

@wdimmit that pr has some breaking changes and intended to be in pair with #60.
@jlkalberer we probably need to make dev branch in spark-protocol.

btw, just tested, the requests issue is gone with the fix.

[jlkalberer]

Yeah, makes sense since people are starting to use this. I'll create a branch.