Hello all,
I just submitted PR #44 for an actionloop based Python Runtime. In my tests, it
is only slightly slower than the top ones (1000 actions in 0.21 against 0.17
for the go and PHP based runtimes).
I tried to make it as backward compatible as possible. It passes all the
existing tests although I had to modify a few of them to adapt to some (IHMO
minor) differences in the runtimes.
Notable differences are:
- at init time it does not log the errors, but returns compilation errors in
the answer, in format {"error": "compilation error"}
- if the action produces at runtime an `{error` it still returns error code
200, not 502
In the first case, it happens because actionloop does not log anything at init
time. I could change it if needed but this way you have a consistent behavior
in actionloop based runtimes.
In the second case, it is a bit tricky. I found a test expecting that when you
raise an exception in the action, I should return an "{"error:"...}" and set
the error code to 502. Unfortunately, because the only communication channel
among the action running a loop and the web server is a pipe, I should parse
the answer (that can be large) just to find it has a key "error" for each
request and I do not want to slow down the runtime just for setting this error
code.
Please revise the PR#44 for the Python runtime. Note also that before all I
need someone merges the PR#4193 on incubator-openwhisk because test data are
placed in the main repository so far and I need a python37_virtualenv.zip there.
