beanstalkd-rs 0.5.1

Packaging, documentation and dependency updates: no protocol, binlog,
cluster or command line changes, so 0.5.0 and 0.5.1 are interchangeable and
a rolling upgrade between them is a plain restart.

Added

• crates.io: the server is published as the bstkd package
  (cargo install bstkd; the binary is still beanstalkd-rs), and
  cargo binstall bstkd installs it too. The package is bstkd because
  beanstalkd-rs on crates.io is
  an unrelated client library. The crates it builds on (bstk-proto,
  bstk-engine, bstk-store, bstk-net, bstk-raft, plus
  bstk-engine-oracle, the frozen engine its tests compare against) are
  published too, in dependency order, by scripts/publish.sh.
• scripts/demo.sh: a demo with no dependencies beyond bash, python3
  and curl. It runs a producer and a worker over the protocol, reads
  /metrics, sends SIGTERM and restarts on the same binlog to show that
  the jobs survived.
• README: a "Migrating from beanstalkd" section (the protocol and the
  command line are compatible; the binlog directory is not, plus the other
  behavior changes), "Is it ready?" (the test evidence) and "When
  something else fits better".
• crates.io metadata for every published crate (description, keywords,
  categories, README) and [package.metadata.binstall] for the server.

Changed

• The server package is named bstkd instead of bstk-server. The binary,
  the release archives, the container image and the systemd unit are still
  beanstalkd-rs.
• Dependencies: rustls 0.23.46, rustls-webpki 0.103.16 and tokio 1.54.1,
  which fix RUSTSEC-2026-0346, -0347, -0382 and -0386.

beanstalkd-rs 0.5.0

First release. A Rust reimplementation of
beanstalkd 1.13 that existing
clients can use unmodified.

Added

• Protocol compatibility: every command, reply, error code and edge case
  of the beanstalkd text protocol, checked byte for byte against the
  reference server by differential tests and real clients (Python
  greenstalk, Go go-beanstalk). The few intentional differences are listed
  in docs/COMPAT.md.
• Reference command line: -l, -p, -z, -b, -f, -F, -s,
  -V, -v with the reference's defaults; --version as an alias of -v.
  -u is rejected: run the server as the desired user (for example with the
  systemd unit in packaging/systemd/).
• Write-ahead log (-b DIR, -f MS, -F, -s BYTES): jobs survive
  restarts and crashes, and no reply is sent before its change reaches the
  log. The on-disk format is not the reference's: a reference binlog
  directory cannot be reused.
• Graceful shutdown on SIGTERM / SIGINT (syncs the log; the reference
  exits abruptly), and drain mode on SIGUSR1 as in the reference.
• Configuration file (--config FILE, TOML; --check-config validates
  it): several listeners, binlog, logging, HTTP and cluster settings. See
  docs/beanstalkd-rs.example.toml and
  the ready-made files in packaging/examples/.
• TLS and mutual TLS listeners, and an optional token authentication
  extension (auth <token>, needs client support) on TLS listeners, with
  limits on unauthenticated connections.
• HTTP monitoring (off by default): /healthz, /readyz, /metrics
  (Prometheus) and /admin (read-only JSON). Log lines carry ANSI colors
  only when stderr is a terminal.
• Cluster mode ([cluster]): 3 or 5 nodes replicate every job and
  connection through Raft over mutual TLS, keep serving through the loss of
  a minority of nodes, and look like one beanstalkd server to clients
  connected to any node. Bootstrap with --cluster-init
  (cluster.initial_voters chooses the first voters); [[cluster.peer]]
  lists seeds and address overrides. A node with an empty data directory
  rejoins against the current voters, one node at a time; a node whose id is
  not a member yet waits until it is added. See
  docs/DESIGN.md
  §8 and docs/OPERATIONS.md.
• Membership changes at runtime with the beanstalkd-rs cluster
  operator command (status, add, promote, remove, set-addr) over
  the cluster port's authenticated admin channel (mTLS with the bstk-admin
  certificate from scripts/mkcluster-certs.sh DIR admin): grow from 1 to 3
  or 3 to 5 nodes, shrink, replace a node or disk, remove the leader, change
  an address, while the cluster serves. Guardrails: one voter added or
  removed per change, only a caught-up learner is promoted, node ids are
  never reused, fewer than 3 voters or a plaintext address off loopback only
  when forced. A removed node stops by itself (exit status 11) once a
  majority of the remaining voters confirm its removal; a node whose Raft
  core stops on a fatal error exits with status 21 so that the service
  manager restarts it.
• Membership metrics on /metrics and /admin (voters, learners,
  members, joint state, learner lag, joining and rejoining phases); cluster
  /readyz is 503 on a node that is isolated, not a member, or still
  starting.
• Operations guide
  (docs/OPERATIONS.md):
  install, configuration, persistence, security, cluster bootstrap, node
  loss and rejoin, membership runbooks, CA rotation, backup and restore,
  upgrades, monitoring and alerts, troubleshooting.
• Performance: 1.2–1.4× the reference's throughput at 10 and 100
  connections with 10–24% less CPU per operation, 2.2× with a binlog at 100
  connections; a 3-node cluster serves about 316k operations per second
  through the leader (tmpfs, Linux on an Apple M6). Worker threads are
  tunable with --threads / server.threads. See
  docs/BENCH.md.
• Packaging: release archives for Linux (x86_64, aarch64; glibc 2.34
  or newer, or fully static with musl) and macOS (aarch64) with SHA-256
  checksums, a container image (Dockerfile), a hardened systemd unit
  (RestartPreventExitStatus=10 11), example configurations
  (packaging/)
  and scripts/mkcluster-certs.sh for a cluster CA and node certificates.