under development
jug.parallel allows processing jug requests in parallel. Under the hood it launches a node.js based load balancer which in turn spins up n different instances of your jug instance. A round-robin approach is used to distribute the requests.
- node.js installation
devtools::install_github("Bart6114/jug.parallel")Start up 8 jug instances in parallel:
library(jug)
jug() %>%
get("/", function(req, res, err){
"Hello World!"
}) %>%
simple_error_handler_json() %>%
serve_it_parallel(processes=8)curl 127.0.0.1:8080Hello World!
One can also call serve_it_parallel with wait=FALSE. This way the terminal will not be blocked and the jug instances will be served in the background. Remember that you will have to clean up the servers manually.
Stop and clean-up all servers:
kill_servers()[1] TRUE
Stopping servers...success
Below (non-exhaustive) load tests have been done using loadtest on a MacBook Pro with a 2,5 GHz Intel Core i7 CPU. The requests per second (rps) parameter was set to 1000.
| type | concurrency | requests | duration (secs) | errors |
|---|---|---|---|---|
| jug | 6 | 1e2 | 0.48 | 0 |
| jug.parallel | 6 | 1e2 | 0.62 | 0 |
| jug | 6 | 1e3 | 1.95 | 1 |
| jug.parallel | 6 | 1e3 | 1.37 | 0 |
| jug | 6 | 1e4 | 15.08 | 458 |
| jug.parallel | 6 | 1e4 | 12.30 | 0 |
The most important take away here is that jug.parallel only becomes interesting in cases where you are expecting a (very) high load or where the request processing duration is significant.