Running the daemon

secantusd-rs --host 127.0.0.1 --port 27017 --storage-path ./secantus-data

The daemon prints secantusd-rs listening on <host>:<port> once the accept loop is up (launchers can parse that line), runs until SIGINT/SIGTERM, and shuts WiredTiger down cleanly on exit. --port 0 binds an OS-assigned port. Bad arguments exit 2; --help / --version exit 0.

CLI flags

Both --flag value and --flag=value spellings work.

Flag	Meaning	Default
--config PATH	Explicit secantusd.toml (see below)	auto-discovered
--host HOST	Bind address	127.0.0.1
--port N	TCP port (0 = OS-assigned)	27017
--storage-path PATH	WiredTiger home directory (created if absent)	./secantus-data
--log-level LEVEL	DEBUG / INFO / WARNING / ERROR	INFO
--cache-size SIZE	WiredTiger cache, unit-suffixed (256M, 1G, 8G)	4G
--session-max N	Max concurrent WiredTiger sessions (≈ client connections)	1000
--sync-on-commit	fsync every commit (closes the j: true gap; large throughput cost)	off
--oplog-async	Persist oplog entries via a background drainer pool instead of inside each write’s commit (higher write throughput; change-stream events surface once drained)	off, or SECANTUS_OPLOG_ASYNC
--oplog-nonlogged	Create oplog tables log=(enabled=false) — checkpoint-durable only; a crash can lose the oplog tail	off, or SECANTUS_OPLOG_NONLOGGED
--data-nonlogged	mongod’s split: WAL-log only the oplog; data tables recover by oplog replay from the last stable checkpoint (see Recovery)	off, or SECANTUS_DATA_NONLOGGED
--checkpoint-seconds N	Seconds between checkpoints. A checkpoint is what trims the log, and under --data-nonlogged what makes the data durable	60, or SECANTUS_CHECKPOINT_SECONDS
--auth	Require SCRAM authentication on every non-handshake command	off
--standalone	Advertise a plain standalone hello instead of the single-node replica-set persona (disables change-stream topology, matters for drivers that gate on isReplicaSet)	off
--noop-heartbeat-seconds N	Periodic oplog no-op so cluster time advances while idle (mongod default is 10s)	0 (disabled)
--oplog-retention-seconds N	Oplog retention window	3600
--oplog-max-entries N	Oplog entry cap (whichever cap hits first prunes oldest)	100000
--oplog-archive-dir PATH	Archive pruned oplog rows for point-in-time recovery beyond the live window	off
--tls-cert-file PATH	Server certificate (PEM); pair with --tls-key-file	off
--tls-key-file PATH	Server private key (PEM)	off
--tls-ca-file PATH	Client-cert CA bundle — turns on mTLS verification	off
--tls-require-client-cert	Reject clients that present no certificate	off

TLS pairing rules are enforced: setting one of cert-file / key-file without the other is a startup error, and the client-cert flags are only meaningful with server TLS configured. See Security.

Slow operations

An operation that runs for 100 ms or longer writes one line to the log, at INFO, as mongod does:

[2026-10-09T19:10:59Z INFO  secantus_server] Slow query {"type":"command","ns":"bench.stall","command":"insert","ctx":"conn3","ninserted":100,"workingMillis":1717,"durationMillis":1717}

The same lines are in the getLog buffer. workingMillis leaves out time an awaitData getMore spent waiting for new data, so an idle change stream logs nothing. A failed operation carries "ok":0 with errCode and errName.

The threshold is the profile command’s slowms, and sampleRate thins the lines. Both are server-wide and last until the server stops:

client.admin.command("profile", 0, slowms=20)   # log anything over 20 ms

The line says which operation was slow and for how long. It does not say where in the storage engine the time went, which mongod’s line does.

Configuration file

Both MongoDB-wire daemons (secantusd-py and secantusd-rs) read the same secantusd.toml, discovered in order:

1. ./secantusd.toml (per-checkout / per-cwd)

2. ~/.secantus/secantusd.toml (per-user)

3. /etc/secantus/secantusd.toml (system-wide)

or passed explicitly with --config. Every key is optional; explicit CLI flags override the file. The sections mirror the flags (a handful of keys are Rust-daemon-only — the commented [storage] modes below — and the Python daemon rejects them, so keep them out of a secantusd.toml shared with secantusd-py):

[server]
host = "127.0.0.1"
port = 27017
storage_path = "./secantus-data"
log_level = "INFO"
auth = false
standalone = false

[oplog]
retention_seconds = 3600.0
max_entries = 100000
noop_heartbeat_seconds = 0.0
# archive_dir = "/var/lib/secantus/pitr-archive"

[storage]
cache_size = "1G"
session_max = 1000
sync_on_commit = false
# Rust-daemon-only storage modes (secantusd-py rejects these keys):
# oplog_async = true
# oplog_nonlogged = true
# data_nonlogged = true
# checkpoint_seconds = 60

# [tls]
# cert_file = "/etc/letsencrypt/live/db.example.com/fullchain.pem"
# key_file  = "/etc/letsencrypt/live/db.example.com/privkey.pem"
# ca_file   = "/etc/ssl/client-ca.crt"
# require_client_cert = false

The annotated example ships as secantusd.toml.example in the repository root. (The legacy secantusdb.toml filename is still auto-discovered at each location.)

Topology persona

By default the server advertises itself in hello as the primary of a single-node secantus replica set — a deliberate fiction that makes drivers’ topology machinery accept change streams. There are no other members and no elections. --standalone switches to a plain standalone hello for tests and drivers that behave differently against replica sets (the Java driver’s getSecondary() helper, for example, spins forever against a replica-set persona with no secondary).