[Q] How does the concurrency aspect actually work?

[fabis94]

RoadRunner advertises high performance through multi-threading and coroutines, but how does this actually work at the PHP level? I couldn't find any resources that explain this in detail, but if there are any I'd love to read them.

Basically, how does RoadRunner ensure that a single PHP app can actually work on multiple requests at the same time, instead of being blocked by a single process at a time like classic PHP?

[rustatian]

A simple diagram is here: #622 (comment)

RR runs a set of processes with PHP workers inside. The workers are similar and path to them specified by the server: https://github.com/spiral/roadrunner-binary/blob/master/.rr.yaml#L23. So, there is no single PHP app, there are a set of processes with a PHP inside and RR manages these processes and extends functionality by the plugins.

You can find simplified docs here: https://github.com/spiral/roadrunner/tree/master/pkg/doc

[rustatian]

@Warxcell So, I guess, that you want smt like that:

# HELP rr_http_worker_status_25727 Worker current state
# TYPE rr_http_worker_status_25727 gauge
rr_http_worker_status_25727{status="ready"} 0
# HELP rr_http_worker_status_25731 Worker current state
# TYPE rr_http_worker_status_25731 gauge
rr_http_worker_status_25731{status="ready"} 0
# HELP rr_http_worker_status_25732 Worker current state
# TYPE rr_http_worker_status_25732 gauge
rr_http_worker_status_25732{status="ready"} 0
# HELP rr_http_worker_status_25734 Worker current state
# TYPE rr_http_worker_status_25734 gauge
rr_http_worker_status_25734{status="ready"} 0
# HELP rr_http_worker_status_25737 Worker current state
# TYPE rr_http_worker_status_25737 gauge
rr_http_worker_status_25737{status="ready"} 0
# HELP rr_http_worker_status_25738 Worker current state
# TYPE rr_http_worker_status_25738 gauge
rr_http_worker_status_25738{status="ready"} 0
# HELP rr_http_worker_status_25739 Worker current state
# TYPE rr_http_worker_status_25739 gauge
rr_http_worker_status_25739{status="ready"} 0
# HELP rr_http_worker_status_25740 Worker current state
# TYPE rr_http_worker_status_25740 gauge
rr_http_worker_status_25740{status="ready"} 0
# HELP rr_http_worker_status_25741 Worker current state
# TYPE rr_http_worker_status_25741 gauge
rr_http_worker_status_25741{status="ready"} 0
# HELP rr_http_worker_status_25742 Worker current state
# TYPE rr_http_worker_status_25742 gauge
rr_http_worker_status_25742{status="ready"} 0
# HELP rr_http_workers_memory_bytes Memory usage by HTTP workers.
# TYPE rr_http_workers_memory_bytes gauge
rr_http_workers_memory_bytes 2.65252864e+08

Where r_http_worker_status_25742 - current worker state (number at the end is PID).
And {status="ready"} is the current worker's state. (working, invalid, killing, destroyed, ready, etc).

[rustatian]

Or even better:

rr_http_worker_memory_bytes{pid="47164"} 2.6013696e+07
rr_http_worker_memory_bytes{pid="47166"} 2.6386432e+07
rr_http_worker_memory_bytes{pid="47168"} 2.7410432e+07
rr_http_worker_memory_bytes{pid="47172"} 2.6349568e+07
rr_http_worker_memory_bytes{pid="47173"} 2.6783744e+07
rr_http_worker_memory_bytes{pid="47174"} 2.7267072e+07
rr_http_worker_memory_bytes{pid="47175"} 2.7066368e+07
rr_http_worker_memory_bytes{pid="47176"} 2.654208e+07
rr_http_worker_memory_bytes{pid="47177"} 2.6914816e+07
rr_http_worker_memory_bytes{pid="47178"} 2.656256e+07
# HELP rr_http_worker_state Worker current state
# TYPE rr_http_worker_state gauge
rr_http_worker_state{pid="47164",state="ready"} 0
rr_http_worker_state{pid="47166",state="ready"} 0
rr_http_worker_state{pid="47168",state="ready"} 0
rr_http_worker_state{pid="47172",state="ready"} 0
rr_http_worker_state{pid="47173",state="ready"} 0
rr_http_worker_state{pid="47174",state="ready"} 0
rr_http_worker_state{pid="47175",state="ready"} 0
rr_http_worker_state{pid="47176",state="ready"} 0
rr_http_worker_state{pid="47177",state="ready"} 0
rr_http_worker_state{pid="47178",state="ready"} 0

[Warxcell]

that would be great, yes. this is how PHPFPM exporter does it:

phpfpm_total_processes{pool="www",scrape_uri="tcp://127.0.0.1:9000/status"} 5
phpfpm_active_processes{pool="www",scrape_uri="tcp://127.0.0.1:9000/status"} 1
phpfpm_idle_processes{pool="www",scrape_uri="tcp://127.0.0.1:9000/status"} 4

• but your variant should be able to aggregate same values.

[rustatian]

Yeah, PHP-FPM has only 3 states, while RR worker has much more. You may track for example workers which are in the stopped state and are prepared to reallocate.
I will add total workers as well.

[rustatian]

So, ok, the final stats will be as following:

# HELP rr_http_total_workers Total number of workers used by the HTTP plugin
# TYPE rr_http_total_workers gauge
rr_http_total_workers 10
# HELP rr_http_worker_memory_bytes Worker current memory usage
# TYPE rr_http_worker_memory_bytes gauge
rr_http_worker_memory_bytes{pid="58137"} 2.7758592e+07
rr_http_worker_memory_bytes{pid="58140"} 2.5870336e+07
rr_http_worker_memory_bytes{pid="58143"} 2.6062848e+07
rr_http_worker_memory_bytes{pid="58146"} 2.648064e+07
rr_http_worker_memory_bytes{pid="58147"} 2.711552e+07
rr_http_worker_memory_bytes{pid="58148"} 2.725888e+07
rr_http_worker_memory_bytes{pid="58149"} 2.67264e+07
rr_http_worker_memory_bytes{pid="58150"} 2.7168768e+07
rr_http_worker_memory_bytes{pid="58151"} 2.7148288e+07
rr_http_worker_memory_bytes{pid="58152"} 2.6959872e+07
# HELP rr_http_worker_state Worker current state
# TYPE rr_http_worker_state gauge
rr_http_worker_state{pid="58137",state="ready"} 0
rr_http_worker_state{pid="58140",state="ready"} 0
rr_http_worker_state{pid="58143",state="ready"} 0
rr_http_worker_state{pid="58146",state="ready"} 0
rr_http_worker_state{pid="58147",state="ready"} 0
rr_http_worker_state{pid="58148",state="ready"} 0
rr_http_worker_state{pid="58149",state="ready"} 0
rr_http_worker_state{pid="58150",state="ready"} 0
rr_http_worker_state{pid="58151",state="ready"} 0
rr_http_worker_state{pid="58152",state="ready"} 0