Improve log streaming

[dminkovsky]

Current mechanism (streaming output from a forked gcloud) seems unreliable. Examine the call gcloud container builds log --stream makes with --log-http and use that with an equivalent HTTP call.

[dminkovsky]

For now using this as a workaround to stream the latest build request:

gcloud container builds list | head -2 | awk 'NR>1{print $1}' | xargs gcloud container builds log --stream

[dminkovsky]

Having examined gcloud container builds log --stream with --log-http, it appears to works like this:

• GET request is made to https://cloudbuild.googleapis.com/v1/projects/PROJECT_NAME/builds/BUILD_ID?alt=json
• Response is examined; probably checks to see whether the status is WORKING
• GET request is made to https://storage.googleapis.com/PROJECT_ID.cloudbuild-logs.googleusercontent.com/log-BUILD_ID.txt. This request includes header Range: bytes=0-.
• Response is printed to terminal up until response's content-length.
• The above two requests are repeated. If the build is still WORKING, then subsequent storage log requests change the Range header's value to offset from already received data. For example. if the first storage log response was content-length: 1650, the next storage log request will be Range: bytes=1650- and so on.

[dminkovsky]

Could also explore forking gcloud container builds log --stream using Node's spawn instead of currently used exec.