mirror of
https://github.com/juspay/hyperswitch.git
synced 2025-11-01 19:42:27 +08:00
395 lines
13 KiB
Rust
395 lines
13 KiB
Rust
use std::{
|
|
sync::{self, atomic},
|
|
time as std_time,
|
|
};
|
|
|
|
use common_utils::errors::CustomResult;
|
|
use diesel_models::enums::{self, ProcessTrackerStatus};
|
|
pub use diesel_models::process_tracker as storage;
|
|
use error_stack::{report, ResultExt};
|
|
use redis_interface::{RedisConnectionPool, RedisEntryId};
|
|
use router_env::{instrument, opentelemetry, tracing};
|
|
use uuid::Uuid;
|
|
|
|
use super::{
|
|
consumer::{self, types::process_data, workflows},
|
|
env::logger,
|
|
};
|
|
use crate::{
|
|
configs::settings::SchedulerSettings, consumer::types::ProcessTrackerBatch, errors,
|
|
flow::SchedulerFlow, metrics, SchedulerAppState, SchedulerInterface,
|
|
};
|
|
|
|
pub async fn divide_and_append_tasks<T>(
|
|
state: &T,
|
|
flow: SchedulerFlow,
|
|
tasks: Vec<storage::ProcessTracker>,
|
|
settings: &SchedulerSettings,
|
|
) -> CustomResult<(), errors::ProcessTrackerError>
|
|
where
|
|
T: SchedulerInterface + Send + Sync + ?Sized,
|
|
{
|
|
let batches = divide(tasks, settings);
|
|
// Safety: Assuming we won't deal with more than `u64::MAX` batches at once
|
|
#[allow(clippy::as_conversions)]
|
|
metrics::BATCHES_CREATED.add(&metrics::CONTEXT, batches.len() as u64, &[]); // Metrics
|
|
for batch in batches {
|
|
let result = update_status_and_append(state, flow, batch).await;
|
|
match result {
|
|
Ok(_) => (),
|
|
Err(error) => logger::error!(error=%error.current_context()),
|
|
}
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
pub async fn update_status_and_append<T>(
|
|
state: &T,
|
|
flow: SchedulerFlow,
|
|
pt_batch: ProcessTrackerBatch,
|
|
) -> CustomResult<(), errors::ProcessTrackerError>
|
|
where
|
|
T: SchedulerInterface + Send + Sync + ?Sized,
|
|
{
|
|
let process_ids: Vec<String> = pt_batch
|
|
.trackers
|
|
.iter()
|
|
.map(|process| process.id.to_owned())
|
|
.collect();
|
|
match flow {
|
|
SchedulerFlow::Producer => {
|
|
state
|
|
.process_tracker_update_process_status_by_ids(
|
|
process_ids,
|
|
storage::ProcessTrackerUpdate::StatusUpdate {
|
|
status: ProcessTrackerStatus::Processing,
|
|
business_status: None,
|
|
},
|
|
)
|
|
.await.map_or_else(|error| {
|
|
logger::error!(error=%error.current_context(),"Error while updating process status");
|
|
Err(error.change_context(errors::ProcessTrackerError::ProcessUpdateFailed))
|
|
}, |count| {
|
|
logger::debug!("Updated status of {count} processes");
|
|
Ok(())
|
|
})
|
|
}
|
|
SchedulerFlow::Cleaner => {
|
|
let res = state
|
|
.reinitialize_limbo_processes(process_ids, common_utils::date_time::now())
|
|
.await;
|
|
match res {
|
|
Ok(count) => {
|
|
logger::debug!("Reinitialized {count} processes");
|
|
Ok(())
|
|
}
|
|
Err(error) => {
|
|
logger::error!(error=%error.current_context(),"Error while reinitializing processes");
|
|
Err(error.change_context(errors::ProcessTrackerError::ProcessUpdateFailed))
|
|
}
|
|
}
|
|
}
|
|
_ => {
|
|
let error_msg = format!("Unexpected scheduler flow {flow:?}");
|
|
logger::error!(error = %error_msg);
|
|
Err(report!(errors::ProcessTrackerError::UnexpectedFlow).attach_printable(error_msg))
|
|
}
|
|
}?;
|
|
|
|
let field_value_pairs = pt_batch.to_redis_field_value_pairs()?;
|
|
|
|
match state
|
|
.stream_append_entry(
|
|
&pt_batch.stream_name,
|
|
&RedisEntryId::AutoGeneratedID,
|
|
field_value_pairs,
|
|
)
|
|
.await
|
|
.change_context(errors::ProcessTrackerError::BatchInsertionFailed)
|
|
{
|
|
Ok(x) => Ok(x),
|
|
Err(mut err) => {
|
|
let update_res = state
|
|
.process_tracker_update_process_status_by_ids(
|
|
pt_batch.trackers.iter().map(|process| process.id.clone()).collect(),
|
|
storage::ProcessTrackerUpdate::StatusUpdate {
|
|
status: ProcessTrackerStatus::Processing,
|
|
business_status: None,
|
|
},
|
|
)
|
|
.await.map_or_else(|error| {
|
|
logger::error!(error=%error.current_context(),"Error while updating process status");
|
|
Err(error.change_context(errors::ProcessTrackerError::ProcessUpdateFailed))
|
|
}, |count| {
|
|
logger::debug!("Updated status of {count} processes");
|
|
Ok(())
|
|
});
|
|
|
|
match update_res {
|
|
Ok(_) => (),
|
|
Err(inner_err) => {
|
|
err.extend_one(inner_err);
|
|
}
|
|
};
|
|
|
|
Err(err)
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn divide(
|
|
tasks: Vec<storage::ProcessTracker>,
|
|
conf: &SchedulerSettings,
|
|
) -> Vec<ProcessTrackerBatch> {
|
|
let now = common_utils::date_time::now();
|
|
let batch_size = conf.producer.batch_size;
|
|
divide_into_batches(batch_size, tasks, now, conf)
|
|
}
|
|
|
|
pub fn divide_into_batches(
|
|
batch_size: usize,
|
|
tasks: Vec<storage::ProcessTracker>,
|
|
batch_creation_time: time::PrimitiveDateTime,
|
|
conf: &SchedulerSettings,
|
|
) -> Vec<ProcessTrackerBatch> {
|
|
let batch_id = Uuid::new_v4().to_string();
|
|
|
|
tasks
|
|
.chunks(batch_size)
|
|
.fold(Vec::new(), |mut batches, item| {
|
|
let batch = ProcessTrackerBatch {
|
|
id: batch_id.clone(),
|
|
group_name: conf.consumer.consumer_group.clone(),
|
|
stream_name: conf.stream.clone(),
|
|
connection_name: String::new(),
|
|
created_time: batch_creation_time,
|
|
rule: String::new(), // is it required?
|
|
trackers: item.to_vec(),
|
|
};
|
|
batches.push(batch);
|
|
|
|
batches
|
|
})
|
|
}
|
|
|
|
pub async fn get_batches(
|
|
conn: &RedisConnectionPool,
|
|
stream_name: &str,
|
|
group_name: &str,
|
|
consumer_name: &str,
|
|
) -> CustomResult<Vec<ProcessTrackerBatch>, errors::ProcessTrackerError> {
|
|
let response = match conn
|
|
.stream_read_with_options(
|
|
stream_name,
|
|
RedisEntryId::UndeliveredEntryID,
|
|
// Update logic for collecting to Vec and flattening, if count > 1 is provided
|
|
Some(1),
|
|
None,
|
|
Some((group_name, consumer_name)),
|
|
)
|
|
.await
|
|
{
|
|
Ok(response) => response,
|
|
Err(error) => {
|
|
if let redis_interface::errors::RedisError::StreamEmptyOrNotAvailable =
|
|
error.current_context()
|
|
{
|
|
logger::debug!("No batches processed as stream is empty");
|
|
return Ok(Vec::new());
|
|
} else {
|
|
return Err(error.change_context(errors::ProcessTrackerError::BatchNotFound));
|
|
}
|
|
}
|
|
};
|
|
|
|
metrics::BATCHES_CONSUMED.add(&metrics::CONTEXT, 1, &[]);
|
|
|
|
let (batches, entry_ids): (Vec<Vec<ProcessTrackerBatch>>, Vec<Vec<String>>) = response.into_values().map(|entries| {
|
|
entries.into_iter().try_fold(
|
|
(Vec::new(), Vec::new()),
|
|
|(mut batches, mut entry_ids), entry| {
|
|
// Redis entry ID
|
|
entry_ids.push(entry.0);
|
|
// Value HashMap
|
|
batches.push(ProcessTrackerBatch::from_redis_stream_entry(entry.1)?);
|
|
|
|
Ok((batches, entry_ids))
|
|
},
|
|
)
|
|
}).collect::<CustomResult<Vec<(Vec<ProcessTrackerBatch>, Vec<String>)>, errors::ProcessTrackerError>>()?
|
|
.into_iter()
|
|
.unzip();
|
|
// Flattening the Vec's since the count provided above is 1. This needs to be updated if a
|
|
// count greater than 1 is provided.
|
|
let batches = batches.into_iter().flatten().collect::<Vec<_>>();
|
|
let entry_ids = entry_ids.into_iter().flatten().collect::<Vec<_>>();
|
|
|
|
conn.stream_acknowledge_entries(stream_name, group_name, entry_ids.clone())
|
|
.await
|
|
.map_err(|error| {
|
|
logger::error!(?error, "Error acknowledging batch in stream");
|
|
error.change_context(errors::ProcessTrackerError::BatchUpdateFailed)
|
|
})?;
|
|
conn.stream_delete_entries(stream_name, entry_ids.clone())
|
|
.await
|
|
.map_err(|error| {
|
|
logger::error!(?error, "Error deleting batch from stream");
|
|
error.change_context(errors::ProcessTrackerError::BatchDeleteFailed)
|
|
})?;
|
|
|
|
Ok(batches)
|
|
}
|
|
|
|
pub fn get_process_tracker_id<'a>(
|
|
runner: storage::ProcessTrackerRunner,
|
|
task_name: &'a str,
|
|
txn_id: &'a str,
|
|
merchant_id: &'a str,
|
|
) -> String {
|
|
format!("{runner}_{task_name}_{txn_id}_{merchant_id}")
|
|
}
|
|
|
|
pub fn get_time_from_delta(delta: Option<i32>) -> Option<time::PrimitiveDateTime> {
|
|
delta.map(|t| common_utils::date_time::now().saturating_add(time::Duration::seconds(t.into())))
|
|
}
|
|
|
|
#[instrument(skip_all)]
|
|
pub async fn consumer_operation_handler<E, T: Send + Sync + 'static>(
|
|
state: T,
|
|
settings: sync::Arc<SchedulerSettings>,
|
|
error_handler_fun: E,
|
|
consumer_operation_counter: sync::Arc<atomic::AtomicU64>,
|
|
workflow_selector: impl workflows::ProcessTrackerWorkflows<T> + 'static + Copy + std::fmt::Debug,
|
|
) where
|
|
// Error handler function
|
|
E: FnOnce(error_stack::Report<errors::ProcessTrackerError>),
|
|
T: SchedulerAppState,
|
|
{
|
|
consumer_operation_counter.fetch_add(1, atomic::Ordering::SeqCst);
|
|
let start_time = std_time::Instant::now();
|
|
|
|
match consumer::consumer_operations(&state, &settings, workflow_selector).await {
|
|
Ok(_) => (),
|
|
Err(err) => error_handler_fun(err),
|
|
}
|
|
let end_time = std_time::Instant::now();
|
|
let duration = end_time.saturating_duration_since(start_time).as_secs_f64();
|
|
logger::debug!("Time taken to execute consumer_operation: {}s", duration);
|
|
|
|
let current_count = consumer_operation_counter.fetch_sub(1, atomic::Ordering::SeqCst);
|
|
logger::info!("Current tasks being executed: {}", current_count);
|
|
}
|
|
|
|
pub fn add_histogram_metrics(
|
|
pickup_time: &time::PrimitiveDateTime,
|
|
task: &mut storage::ProcessTracker,
|
|
stream_name: &str,
|
|
) {
|
|
#[warn(clippy::option_map_unit_fn)]
|
|
if let Some((schedule_time, runner)) = task.schedule_time.as_ref().zip(task.runner.as_ref()) {
|
|
let pickup_schedule_delta = (*pickup_time - *schedule_time).as_seconds_f64();
|
|
logger::error!(%pickup_schedule_delta, "<- Time delta for scheduled tasks");
|
|
let runner_name = runner.clone();
|
|
metrics::CONSUMER_STATS.record(
|
|
&metrics::CONTEXT,
|
|
pickup_schedule_delta,
|
|
&[opentelemetry::KeyValue::new(
|
|
stream_name.to_owned(),
|
|
runner_name,
|
|
)],
|
|
);
|
|
};
|
|
}
|
|
|
|
pub fn get_schedule_time(
|
|
mapping: process_data::ConnectorPTMapping,
|
|
merchant_name: &str,
|
|
retry_count: i32,
|
|
) -> Option<i32> {
|
|
let mapping = match mapping.custom_merchant_mapping.get(merchant_name) {
|
|
Some(map) => map.clone(),
|
|
None => mapping.default_mapping,
|
|
};
|
|
|
|
// For first try, get the `start_after` time
|
|
if retry_count == 0 {
|
|
Some(mapping.start_after)
|
|
} else {
|
|
get_delay(
|
|
retry_count,
|
|
mapping.count.iter().zip(mapping.frequency.iter()),
|
|
)
|
|
}
|
|
}
|
|
|
|
pub fn get_pm_schedule_time(
|
|
mapping: process_data::PaymentMethodsPTMapping,
|
|
pm: &enums::PaymentMethod,
|
|
retry_count: i32,
|
|
) -> Option<i32> {
|
|
let mapping = match mapping.custom_pm_mapping.get(pm) {
|
|
Some(map) => map.clone(),
|
|
None => mapping.default_mapping,
|
|
};
|
|
|
|
if retry_count == 0 {
|
|
Some(mapping.start_after)
|
|
} else {
|
|
get_delay(
|
|
retry_count,
|
|
mapping.count.iter().zip(mapping.frequency.iter()),
|
|
)
|
|
}
|
|
}
|
|
|
|
/// Get the delay based on the retry count
|
|
fn get_delay<'a>(
|
|
retry_count: i32,
|
|
mut array: impl Iterator<Item = (&'a i32, &'a i32)>,
|
|
) -> Option<i32> {
|
|
match array.next() {
|
|
Some(ele) => {
|
|
let v = retry_count - ele.0;
|
|
if v <= 0 {
|
|
Some(*ele.1)
|
|
} else {
|
|
get_delay(v, array)
|
|
}
|
|
}
|
|
None => None,
|
|
}
|
|
}
|
|
|
|
pub(crate) async fn lock_acquire_release<T, F, Fut>(
|
|
state: &T,
|
|
settings: &SchedulerSettings,
|
|
callback: F,
|
|
) -> CustomResult<(), errors::ProcessTrackerError>
|
|
where
|
|
F: Fn() -> Fut,
|
|
T: SchedulerInterface + Send + Sync + ?Sized,
|
|
Fut: futures::Future<Output = CustomResult<(), errors::ProcessTrackerError>>,
|
|
{
|
|
let tag = "PRODUCER_LOCK";
|
|
let lock_key = &settings.producer.lock_key;
|
|
let lock_val = "LOCKED";
|
|
let ttl = settings.producer.lock_ttl;
|
|
|
|
if state
|
|
.acquire_pt_lock(tag, lock_key, lock_val, ttl)
|
|
.await
|
|
.change_context(errors::ProcessTrackerError::ERedisError(
|
|
errors::RedisError::RedisConnectionError.into(),
|
|
))?
|
|
{
|
|
let result = callback().await;
|
|
state
|
|
.release_pt_lock(tag, lock_key)
|
|
.await
|
|
.map_err(errors::ProcessTrackerError::ERedisError)?;
|
|
result
|
|
} else {
|
|
Ok(())
|
|
}
|
|
}
|