elixir/rel/vm.args.eex

Rama main · réplica de lectura del repositorio original.

# Customize flags given to the VM: http://erlang.org/doc/man/erl.html
# -mode/-name/-sname/-setcookie are configured via env vars, do not set them here

# Erlang distribution tick time - controls dead node detection speed.
#
# Erlang sends "ticks" between connected nodes. If 4 consecutive ticks are
# missed, the node is considered dead and disconnected. With the default
# net_ticktime of 60s, dead nodes take ~4 minutes to detect.
#
# We reduce this to 15s so dead nodes are detected in ~60 seconds. This is
# important during rolling deployments when VMs are terminated - faster
# detection means the cluster stabilizes more quickly and avoids prolonged
# partition detection by the `global` module.
#
# Trade-off: More sensitive to transient network issues. 15s is still
# conservative enough for cloud environments with occasional packet loss.
-kernel net_ticktime 15

# Number of dirty schedulers doing IO work (file, sockets, and others)
#
# Interacting with the file system usually goes through the async pool.
# Increasing the pool increasing boot time but it will
# likely increase performance for the plug static layer.
+SDio 20

# Double the default maximum ports value
# +Q 131072

# Bind schedulers to CPU's (good when there are no other processes in OS that bind to processors)
# +stbt db

# Disable schedulers compaction of load (don't disable schedulers that is out of work)
#
# This is good for latency and also will keep our Google sustainable load discount higher.
# +scl false

# Enable port parallelism (good for parallelism, bad for latency)
# +spp true

# Doubles the distribution buffer busy limit (good for latency, increases memory consumption)
# +zdbbl 2048

# Tweak GC to run more often
#-env ERL_FULLSWEEP_AFTER 10