Configuration Reference

From ANTFARM Wiki

All ANTFARM services read from /etc/antfarm/. Files are simple key = value under a section header. Comments start with #. Booleans are true/false. Durations are seconds unless the key ends in _ms.

Single-host development config: 📎 antfarm-singlehost.conf.

[edit] queen

[queen]
name            = queen01
listen          = 0.0.0.0:7420
bus             = phero-a.hm.internal:7421
peer            = queen02.hm.internal:7420
epoch_file      = /var/lib/antfarm/epoch
assign_timeout  = 300        # reissue after this many seconds
hb_interval     = 10
hb_miss_evict   = 3
election_min_ms = 1500       # see Queen election notes (not on this wiki)
hosts_file      = /etc/antfarm/hosts.yaml

assign_timeout is the value that makes AF-44 happen. Lowering it makes it happen more.

[edit] forager

[forager]
name            = forager-07
queen           = queen01.hm.internal:7420
bus             = phero-a.hm.internal:7421
slots           = 2
capabilities    = fetch,convert,index,model.shard0
cm_cache_ttl    = 120
manifest_cache  = /var/cache/antfarm/manifest.json
model.strict    = false      # true: missing loader is an error, not a fallback
model.conf      = /etc/antfarm/model.conf

[edit] phero

[phero]
listen          = 0.0.0.0:7421
data_dir        = /var/lib/phero
trail_log_size  = 10000      # messages kept per trail for replay
trail_log_bytes = 0          # 0 = unlimited. see AF-70.
stats_listen    = 127.0.0.1:7421   # sic — same port, different bind, works by accident

[edit] cmsync

[cmsync]
listen          = 0.0.0.0:7422
peer            = queen02.hm.internal:7422
sync_interval   = 2
full_resync_after = 60
snapshot_dir    = /var/backups/cmsync

[edit] toolreg

[toolreg]
listen          = 0.0.0.0:7423
manifest        = /etc/antfarm/manifest.json
schema          = toolreg/2

[edit] antc

antc has no config file. It reads ANTC_MANIFEST from the environment and defaults to http://toolreg.hm.internal:7423/manifest.json.

[edit] inference (3.0+)

[inference]
split           = 1          # shards; see Distributed_Inference_Notes
shard_hosts     = forager-07,forager-09,forager-12,forager-14
warmup_wait     = true
act_encoding    = base64     # the only value that was ever implemented

[edit] Reloading

SIGHUP reloads queen, forager, toolreg. phero and cmsync need a restart; phero keeps its replay log across restarts, cmsync does a full resync on start.

[edit] See also

• Worker Node Registry
• Orchestrator Recovery
• Getting Started