can't create rstudio-server-gpu VM

[goldingn]

I have authorised GPU quota on cloud compute, but trying to initialise a vm with the "rstudio-server-gpu" image errors:

rss_gpu <- gce_vm(template = "rstudio",
                  name = "rstudio-server-gpu",
                  username = "nick",
                  password = "",
                  predefined_type = "n1-standard-16",
                  acceleratorCount = 2L)

2017-09-15 11:30:05> Creating template VM
Request Status Code: 400
Error in checkGoogleAPIError(req) : 
  JSON fetch error: Instances with guest accelerators do not support live migration.

I can create and restart CPU instances just fine, and manually create a GPU instance via the console.

Any ideas how to fix this?

I'm using version 0.1.0.9002 (cloudyr fork, master branch, commit 8d6531b) and this is my session info: ``` R version 3.4.1 (2017-06-30) Platform: x86_64-apple-darwin16.7.0 (64-bit) Running under: macOS Sierra 10.12.6

Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libLAPACK.dylib

locale:
[1] en_GB.UTF-8/en_GB.UTF-8/en_GB.UTF-8/C/en_GB.UTF-8/en_GB.UTF-8

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] googleComputeEngineR_0.1.0.9002

loaded via a namespace (and not attached):
[1] Rcpp_0.12.12 rstudioapi_0.6.0.9000 magrittr_1.5 devtools_1.13.3 xtable_1.8-2
[6] R6_2.2.2 httr_1.3.1 globals_0.10.2 tools_3.4.1 parallel_3.4.1
[11] googleAuthR_0.5.1 withr_2.0.0 tfruns_0.9.1 htmltools_0.3.6 openssl_0.9.7
[16] yaml_2.1.14 assertthat_0.2.0 digest_0.6.12 crayon_1.3.2.9000 tensorflow_1.3.1
[21] shiny_1.0.5 codetools_0.2-15 testthat_1.0.2.9000 curl_2.8.1 memoise_1.1.0
[26] mime_0.5 compiler_3.4.1 reticulate_1.1 jsonlite_1.5 future_1.6.1
[31] httpuv_1.3.5 listenv_0.6.0

</details>

[MarkEdmondson1234]

Its a weird error, but I think you need to add the use_beta=TRUE flag. Or perhaps GPUs are only supported in certain zones.

Note its very experimental.

[goldingn]

Adding use_beta=TRUE, I still get the error (plus a little warning):

2017-09-18 11:18:06> Creating template VM
Request Status Code: 400
Error in checkGoogleAPIError(req) : 
  JSON fetch error: Instances with guest accelerators do not support live migration.
In addition: Warning message:
In (function (name, predefined_type = "f1-micro", image_project = "debian-cloud",  :
  This is using the beta version of the Google Compute Engine API and may not work in the future.

GPUs are definitely available in asia-east-a (which I'm using) though perhaps they are using a different API there. Which zone have you run these on?

Yeah, I appreciate this is experimental (as is Google's API for GPUs). Thanks for your work on this - it'll be incredibly useful!

[MarkEdmondson1234]

Does this occur for new instances or for an existing one? It may be that the zone you launched a previous GPU enabled VM has had a maintenance window, and it got closed down?

I've been launching them in US zones.

[goldingn]

It occurs when creating new instances. I can connect to a GPU-enabled instance (created via the cloud console web app), and I can create a non-GPU enabled instance, but I can't create a GPU-enabled instance.

This section of the Google cloud compute docs seems relevant, though I have no idea how to configure those settings!

[goldingn]

Oh, this looks like the relevant config information

[MarkEdmondson1234]

Ok, it looks like a config flag is on for GPU instances that shouldn't be, will check.

[MarkEdmondson1234]

This works via the code in #101 and will look to improve the template so its much easier