Hi all,
taking this discussion from Slack to here:
What would be the best way to support long running actions, especially actions
for which the duration is unknown when the action is started.
An example is e.g. a blackbox container that transcodes a video. The main issue
is that OW limits the a
Just a quick thought ... May be do it like its done in Travis
1. Each action gets an upper limit say 1 hr (its 50 min in Travis)
2. If there is no progress logging done for say 10 mins then task gets
killed. Here callbacks or long polling would be better
3. An action can still have multiple sub
One of most frequent complaints[1][2][3] I hear from developers using
serverless platforms is coping with cold-start latency when dealing with
sudden bursts of traffic.
Developers often ask for a feature where they can set the number of warm
containers kept in the cache for a function. This would
for nodejs at least: the cost of a few requires of common packages can
easily get you up to the 150-200ms range (e.g. request is a big hitter; and
this is all on top of the cost of starting a container!). perhaps, for
nodejs at least, there are only a few options, ultimately: user pays more
for idl
Speaking to external developers about this, people seem happy to pay for
this feature.
On 31 May 2018 at 12:34, Nick Mitchell wrote:
> for nodejs at least: the cost of a few requires of common packages can
> easily get you up to the 150-200ms range (e.g. request is a big hitter; and
> this is al
Well, Go actions can be precompiled using the same docker image used for
execution, so if you can use Go for implementing your action, you have the
"prewarming" feature out of the box.
Also since Go programs can embed statics, you do not have to import anything,
it is all in memory.
I am tr
The recent work introduced also support to use docker image as a compiler (and
it is what I normally do).
I "compile" the binary offliine and I send it as a single binary, that is
started at init time.
However while for Go and Swift it is "easy", I do not see how we can
"precompile" a nodejs a
Acknowledging that dev cycles, performance tests, or other spiky traffic
may not be initially captured, I would put my $$$ on teaching OW how to
learn the traffic pattern and let it pre-warm based on that. There are
existing algorithms that could be used for prediction, if we just have the
features
