Migrate from thin to puma for the web server

[j05h]

What? Why?

Thin still does not support Rack 3.0, which is needed because Sinatra needs an update due to GHSA-hxx2-7vcw-mqr3.

Given the Thin update is almost a year old, this PR migrates from Thin to Puma as the server for metrics.

How was it tested?

Integrated with a Rails app:

❯ curl 0.0.0.0:9466/metrics
# HELP ruby_collector_metrics_total Total metrics processed by exporter.
# TYPE ruby_collector_metrics_total counter
ruby_collector_metrics_total 4


# HELP ruby_collector_sessions_total Total send_metric sessions processed by exporter.
# TYPE ruby_collector_sessions_total counter
ruby_collector_sessions_total 4


# HELP ruby_collector_bad_metrics_total Total mis-handled metrics by collector.
# TYPE ruby_collector_bad_metrics_total counter



# HELP ruby_collector_working Is the main process collector able to collect metrics
# TYPE ruby_collector_working gauge
ruby_collector_working 1


# HELP ruby_collector_rss total memory used by collector process
# TYPE ruby_collector_rss gauge
ruby_collector_rss 127296

# HELP ruby_heap_free_slots Free ruby heap slots.
# TYPE ruby_heap_free_slots gauge
ruby_heap_free_slots{type="web",pid="12741"} 48754

# HELP ruby_heap_live_slots Used ruby heap slots.
# TYPE ruby_heap_live_slots gauge
ruby_heap_live_slots{type="web",pid="12741"} 477767

# HELP ruby_rss Total RSS used by process.
# TYPE ruby_rss gauge
ruby_rss{type="web",pid="12741"} 0

# HELP ruby_major_gc_ops_total Major GC operations by process.
# TYPE ruby_major_gc_ops_total counter
ruby_major_gc_ops_total{type="web",pid="12741"} 9

# HELP ruby_minor_gc_ops_total Minor GC operations by process.
# TYPE ruby_minor_gc_ops_total counter
ruby_minor_gc_ops_total{type="web",pid="12741"} 56

# HELP ruby_allocated_objects_total Total number of allocated objects by process.
# TYPE ruby_allocated_objects_total counter
ruby_allocated_objects_total{type="web",pid="12741"} 3003928

# HELP ruby_puma_workers_total Number of puma workers.
# TYPE ruby_puma_workers_total gauge
ruby_puma_workers_total{phase="0"} 3

# HELP ruby_puma_booted_workers_total Number of puma workers booted.
# TYPE ruby_puma_booted_workers_total gauge
ruby_puma_booted_workers_total{phase="0"} 0

# HELP ruby_puma_old_workers_total Number of old puma workers.
# TYPE ruby_puma_old_workers_total gauge
ruby_puma_old_workers_total{phase="0"} 0

Resident set size (RSS) graph for a simple Rails app. 14:45-16:00 is the period we had this deployed (along with an upgrade from Rails 7.0 to 7.1). The memory footprint does grow, but only by about 3mb.

[splittingred]

Looks good. Any impact to memory usage relative to thin?

[splittingred]

Also @j05h - does this cause any issue when being used alongside another Puma server in a separate thread?

[j05h]

Also @j05h - does this cause any issue when being used alongside another Puma server in a separate thread?

I haven't seen any issues to date. We've upgraded an app with Rails 7.1 and are doing some testing on that now.

[j05h]

Looks good. Any impact to memory usage relative to thin?

They seem relatively equivalent.

On thin:

❯ top -l 1 | grep 68335 | awk '{print "MEM="$8 "\tCMPRS="$10}'
MEM=62M	CMPRS=59M
MEM=62M	CMPRS=59M
MEM=62M	CMPRS=59M
MEM=2401K	CMPRS=2032K
MEM=2657K	CMPRS=2288K
MEM=168M	CMPRS=110M

On Puma:

❯ top -l 1 | grep "60714" | awk '{print "MEM="$8 "\tCMPRS="$10}'
MEM=55M	CMPRS=53M
MEM=52M	CMPRS=49M
MEM=52M	CMPRS=50M
MEM=2305K	CMPRS=1920K
MEM=2769K	CMPRS=2384K
MEM=165M	CMPRS=158M

This is local on my Mac, so likely its not an Apples-to-Penguins comparison to Linux. Will try to get some data from our test app.