James Andariese
bd00e26bde
fixes: should attempt to create bucket if it doesn't exist #4 must _run_ services, not just say it is. #3 improve cli ergonomics #2 WaitCR does not renew lock in between R intervals (and should) #1
370 lines
13 KiB
Rust
370 lines
13 KiB
Rust
/*
|
|
Copyright 2024 James Andariese
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
use anyhow::{anyhow, Result};
|
|
use bytes::Bytes;
|
|
use clap::{Parser};
|
|
use clap_verbosity_flag::{Verbosity, WarnLevel};
|
|
use tokio::process::{Command};
|
|
use tokio::time::{sleep, Duration};
|
|
|
|
use async_nats::{jetstream};
|
|
|
|
|
|
#[derive(Parser)]
|
|
struct Cli {
|
|
#[command(flatten)]
|
|
verbosity: Verbosity<WarnLevel>,
|
|
|
|
#[arg(long)]
|
|
/// use NATS, via this server. may be any valid nats:// address
|
|
nats: String,
|
|
#[arg(long)]
|
|
/// NATS: kv store bucket
|
|
bucket: String,
|
|
#[arg(long)]
|
|
/// NATS: kv store key
|
|
key: String,
|
|
#[arg(long)]
|
|
/// unique identifier for this agent.
|
|
/// for host-based clusters, hostname is usually a good choice.
|
|
token: String,
|
|
#[arg(long)]
|
|
/// script to run to check the health of the service;
|
|
/// if passed `active`, validate actual health;
|
|
/// if passed `standby`, validate candidacy for service
|
|
healthcheck: String,
|
|
#[arg(long)]
|
|
/// script to assert the started state of the service
|
|
start: String,
|
|
#[arg(long)]
|
|
/// script to assert the stopped state of the service
|
|
stop: String,
|
|
|
|
#[arg(short, default_value_t = 3000)]
|
|
/// renewal interval, the length of time between health checks
|
|
r: u64,
|
|
#[arg(short, default_value_t = 3)]
|
|
/// failures, r*f is when another agent will attempt to take the lock
|
|
f: u64,
|
|
#[arg(short, default_value_t = 2)]
|
|
/// confirms, r*c is when a new agent will begin fencing and services
|
|
c: u64,
|
|
}
|
|
|
|
struct Invocation {
|
|
args: Cli,
|
|
js: Option<jetstream::Context>,
|
|
latest_revision: u64,
|
|
latest_value: Bytes,
|
|
starts: u64,
|
|
exiting: bool,
|
|
active: bool,
|
|
}
|
|
|
|
impl Invocation {
|
|
async fn new() -> Result<Self> {
|
|
let mut r = Invocation {
|
|
args: Cli::parse(),
|
|
js: None,
|
|
latest_revision: 0,
|
|
latest_value: "".into(),
|
|
starts: 0,
|
|
exiting: false,
|
|
active: false,
|
|
};
|
|
r.setup().await?;
|
|
Ok(r)
|
|
}
|
|
|
|
async fn get_store(&mut self) -> Result<jetstream::kv::Store> {
|
|
let js = self.js.clone().ok_or(anyhow!("Jetstream not connected"))?;
|
|
let store_r = js.get_key_value(&self.args.bucket).await;
|
|
if let Ok(store) = store_r {
|
|
return Ok(store);
|
|
}
|
|
let bucket = self.args.bucket.clone();
|
|
log::warn!("bucket {bucket} does not exist. attempting to create it");
|
|
// it's either gonna work or it's not. yolooooooooooooooo
|
|
return Ok(
|
|
js.create_key_value(jetstream::kv::Config {
|
|
bucket: self.args.bucket.clone(),
|
|
..Default::default()
|
|
}).await?
|
|
);
|
|
}
|
|
|
|
async fn update_lock_data(&mut self) -> Result<()> {
|
|
let store = self.get_store().await?;
|
|
let e = store.entry(&self.args.key).await?;
|
|
match e {
|
|
Some(k) => {
|
|
self.latest_value = k.value.into();
|
|
self.latest_revision = k.revision;
|
|
}
|
|
None => {
|
|
self.latest_value = "".into();
|
|
self.latest_revision = 0;
|
|
}
|
|
};
|
|
Ok(())
|
|
}
|
|
|
|
async fn clear_lock(&mut self) -> Result<()> {
|
|
let store = self.get_store().await?;
|
|
store.update(&self.args.key, "".into(), self.latest_revision).await?;
|
|
self.update_lock_data().await?;
|
|
Ok(())
|
|
}
|
|
|
|
async fn take_lock(&mut self) -> Result<()> {
|
|
let store = self.get_store().await?;
|
|
store.update(&self.args.key, self.args.token.clone().into(), self.latest_revision).await?;
|
|
self.update_lock_data().await?;
|
|
Ok(())
|
|
}
|
|
|
|
async fn setup(&mut self) -> Result<()> {
|
|
let nc = async_nats::connect(&self.args.nats).await?;
|
|
println!("info: {:?}", nc.server_info());
|
|
self.js = Some(async_nats::jetstream::new(nc));
|
|
|
|
if self.args.verbosity.is_present() {
|
|
env_logger::Builder::new().filter_level(self.args.verbosity.log_level_filter()).init();
|
|
} else {
|
|
env_logger::Builder::from_env(env_logger::Env::default().default_filter_or("warn")).init();
|
|
}
|
|
|
|
log::error!("error level active");
|
|
log::warn!("warn level active");
|
|
log::info!("info level active");
|
|
log::debug!("debug level active");
|
|
log::trace!("trace level active");
|
|
Ok(())
|
|
}
|
|
|
|
async fn pump(&mut self) -> Result<()> {
|
|
self.update_lock_data().await?;
|
|
if self.latest_value == "" || self.latest_value == self.args.token { // lk empty or mine?
|
|
log::trace!("lk empty or mine? - yes - Found an empty lock or my own token present");
|
|
|
|
if self.healthcheck().await? {
|
|
log::trace!("health check succeeded");
|
|
let max_i = self.args.f;
|
|
for i in 1..=max_i {
|
|
if self.take_lock().await.is_ok() {
|
|
self.unfence_and_enable_process().await?;
|
|
log::trace!("successfully unfenced and enabled process. ending pump cycle.");
|
|
return Ok(());
|
|
} else {
|
|
log::trace!("failed to take lock. attempt {i}/{max_i}");
|
|
}
|
|
}
|
|
} else {
|
|
log::info!("sunny side health check failed");
|
|
}
|
|
// lock failed to write or health check failed at this point
|
|
log::trace!("attempting to clear lock and kill process");
|
|
match self.kill_process().await {
|
|
Ok(()) => {
|
|
log::info!("killed process");
|
|
}
|
|
e => {
|
|
log::info!("failed to kill process: {e:?}");
|
|
}
|
|
}
|
|
match self.clear_lock().await {
|
|
Ok(()) => {
|
|
log::info!("cleared lock");
|
|
}
|
|
e => {
|
|
log::info!("failed to clear lock (will not retry but will wait 1R extra): {e:?}");
|
|
self.wait_r(1).await;
|
|
}
|
|
}
|
|
log::trace!("waiting F*R");
|
|
self.wait_r(self.args.f).await;
|
|
return Ok(());
|
|
} else {
|
|
log::trace!("lk empty or mine? - no - Found a different token in the lock");
|
|
self.kill_process().await?;
|
|
log::trace!("kill_process asserted");
|
|
if !self.healthcheck().await? {
|
|
log::trace!("initial takeover health check failed. not eligible for takeover. ending pump cycle.");
|
|
return Ok(());
|
|
}
|
|
|
|
log::trace!("initial takeover health check succeeded. waiting F for takeover confirmation.");
|
|
self.wait_r(self.args.f).await;
|
|
if !self.healthcheck().await? {
|
|
log::trace!("confirmation takeover health check failed. no longer eligible for takeover. ending pump cycle.");
|
|
return Ok(());
|
|
}
|
|
log::trace!("confirmation takeover health check succeeded.");
|
|
if self.take_lock().await.is_ok() {
|
|
log::info!("lock taken. waiting C for other services to die.");
|
|
for _ in 1..=self.args.c {
|
|
self.wait_r(1).await;
|
|
self.take_lock().await?;
|
|
log::trace!("lock asserted while waiting for C");
|
|
// if this one causes us to exit early, it will simply cause an additional
|
|
// wait of C*R but without lock assertions so it could jeopardize the
|
|
// promotion of this agent. This agent which just failed to assert the
|
|
// lock for some reason...
|
|
// seems legit.
|
|
}
|
|
} else {
|
|
// this happens on every cycle whether the lock _needs_ to be taken or not so
|
|
// logging it every time is not helpful.
|
|
log::trace!("failed to take lock. ending cycle");
|
|
return Ok(());
|
|
}
|
|
}
|
|
log::trace!("ending pump cycle normally.");
|
|
Ok(())
|
|
}
|
|
|
|
async fn run(&mut self) -> Result<()> {
|
|
let mut round = 0;
|
|
self.starts = self.starts + 1;
|
|
let this_start = self.starts;
|
|
|
|
// we must wait C*R before participating in case we had just
|
|
// taken the lock and the program crashed and then restarted
|
|
// within .00001s because this is the year 2050 and computers
|
|
// are really fast now. This is because we must guarantee
|
|
// that the previous owner agent has C*R to disable itself.
|
|
// we can guarantee at all other times that we've waited the
|
|
// requisite time except in circumstances where we have left
|
|
// the control of the round_timer loop.
|
|
//
|
|
// For the same reason, this is also done whenever pump has
|
|
// failed. See below.
|
|
self.wait_r(self.args.c).await;
|
|
|
|
loop {
|
|
round = round + 1;
|
|
if self.exiting {
|
|
log::info!("exiting before round {round}");
|
|
break;
|
|
}
|
|
|
|
// BEWARE vvv
|
|
// this must be awaited before the loop ends.
|
|
let round_timer = sleep(Duration::from_millis(self.args.r));
|
|
// BEWARE ^^^
|
|
|
|
match self.pump().await {
|
|
Ok(()) => {
|
|
log::info!("pump {this_start}.{round}: success");
|
|
}
|
|
e => {
|
|
// See longer comment above about C*R
|
|
self.wait_r(self.args.c).await;
|
|
log::error!("pump {this_start}.{round}: error: {e:?}");
|
|
}
|
|
}
|
|
|
|
// BEWARL PART DEUX
|
|
round_timer.await;
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
async fn kill_process(&mut self) -> Result<()> {
|
|
log::trace!("attempting to kill process");
|
|
let was_active = self.active;
|
|
self.active = false;
|
|
|
|
if was_active {
|
|
log::warn!("killing process");
|
|
}
|
|
let started = Command::new("/bin/sh")
|
|
.arg("-c")
|
|
.arg(self.args.stop.clone())
|
|
.arg("kill")
|
|
.spawn()?.wait().await?.success();
|
|
if started {
|
|
log::trace!("process killed successfully");
|
|
if !self.active {
|
|
log::warn!("process killed successfully");
|
|
}
|
|
} else {
|
|
log::warn!("kill process failed");
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
async fn unfence_and_enable_process(&mut self) -> Result<()> {
|
|
log::trace!("attempting to unfence and enable process");
|
|
let was_active = self.active;
|
|
self.active = true;
|
|
if !was_active {
|
|
log::warn!("starting process");
|
|
}
|
|
let started = Command::new("/bin/sh")
|
|
.arg("-c")
|
|
.arg(self.args.start.clone())
|
|
.arg("start")
|
|
.arg(if self.active {"active"} else {"standby"})
|
|
.spawn()?.wait().await?.success();
|
|
if started {
|
|
log::trace!("process started successfully");
|
|
if !self.active {
|
|
log::warn!("process started successfully");
|
|
}
|
|
} else {
|
|
log::warn!("unfence/enable process failed");
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
async fn healthcheck(&mut self) -> Result<bool> {
|
|
let mut child = Command::new("/bin/sh")
|
|
.arg("-c")
|
|
.arg(self.args.healthcheck.clone())
|
|
.arg("healthcheck")
|
|
.arg(if self.active {"active"} else {"standby"})
|
|
.spawn()?;
|
|
tokio::select! {
|
|
status = child.wait() => {
|
|
Ok(status?.success())
|
|
}
|
|
_ = sleep(Duration::from_millis(self.args.r * self.args.f)) => {
|
|
log::warn!("health check timed out.");
|
|
Ok(false)
|
|
}
|
|
}
|
|
}
|
|
|
|
async fn wait_r(&mut self, count: u64) {
|
|
sleep(Duration::from_millis(self.args.r * count)).await;
|
|
}
|
|
}
|
|
|
|
#[tokio::main]
|
|
async fn main() -> Result<()> {
|
|
log::info!("here we come");
|
|
let mut inv = Invocation::new().await?;
|
|
|
|
log::info!("here we are");
|
|
// inv.check().await?;
|
|
let rr = inv.run().await;
|
|
log::info!("{rr:?}");
|
|
|
|
Ok(())
|
|
}
|