feat(exporter): adding pool alert metrics

Signed-off-by: Abhilash Shetty <abhilash.shetty@datacore.com>

Author: abhilashshetty04

metrics-exporter/src/bin/io_engine/client/pool.rs

@@ -10,6 +10,16 @@ pub(crate) struct PoolInfo {
     committed: u64,
     disk_capacity: u64,
     max_expandable_size: u64,
+    io_error_count: u64,
+    io_error_threshold: u64,
+    io_stalled: bool,
+    io_stall_transition_count: u64,
+    io_stall_transition_threshold: u64,
+    alert_status: i32,
+    notice: Vec<i32>,
+    attention: Vec<i32>,
+    warning: Vec<i32>,
+    critical: Vec<i32>,
 }
 
 impl PoolInfo {
@@ -43,10 +53,60 @@ impl PoolInfo {
         self.max_expandable_size
     }
 
-    /// Get pool of the io_engine.
+    /// Get state of the Pool.
     pub(crate) fn state(&self) -> u64 {
         self.state
     }
+
+    /// Get the count of IO errors on the pool.
+    pub(crate) fn io_error_count(&self) -> u64 {
+        self.io_error_count
+    }
+
+    /// Get the IO error threshold for the pool.
+    pub(crate) fn io_error_threshold(&self) -> u64 {
+        self.io_error_threshold
+    }
+
+    /// Get whether the pool is currently in stalled state.
+    pub(crate) fn io_stalled(&self) -> bool {
+        self.io_stalled
+    }
+
+    /// Get the count of IO stall transitions on the pool.
+    pub(crate) fn io_stall_transition_count(&self) -> u64 {
+        self.io_stall_transition_count
+    }
+
+    /// Get the IO stall transition threshold for the pool.
+    pub(crate) fn io_stall_transition_threshold(&self) -> u64 {
+        self.io_stall_transition_threshold
+    }
+
+    /// Get the alert status for the pool.
+    pub(crate) fn alert_status(&self) -> i32 {
+        self.alert_status
+    }
+
+    /// Get the collection of notice alert reasons for the pool.
+    pub(crate) fn notice(&self) -> &Vec<i32> {
+        &self.notice
+    }
+
+    /// Get the collection of attention alert reasons for the pool.
+    pub(crate) fn attention(&self) -> &Vec<i32> {
+        &self.attention
+    }
+
+    /// Get the collection of warning alert reasons for the pool.
+    pub(crate) fn warning(&self) -> &Vec<i32> {
+        &self.warning
+    }
+
+    /// Get the collection of critical alert reasons for the pool.
+    pub(crate) fn critical(&self) -> &Vec<i32> {
+        &self.critical
+    }
 }
 
 /// Array of PoolInfo objects.
@@ -57,6 +117,34 @@ pub(crate) struct Pools {
 
 impl From<rpc::v1::pool::Pool> for PoolInfo {
     fn from(value: rpc::v1::pool::Pool) -> Self {
+        let mut io_error_count: u64 = 0;
+        let mut io_error_threshold: u64 = 0;
+        let mut io_stalled: bool = false;
+        let mut io_stall_transition_count: u64 = 0;
+        let mut io_stall_transition_threshold: u64 = 0;
+        let mut alert_status: i32 = 0;
+        let mut notice: Vec<i32> = Vec::new();
+        let mut attention: Vec<i32> = Vec::new();
+        let mut warning: Vec<i32> = Vec::new();
+        let mut critical: Vec<i32> = Vec::new();
+        if let Some(errors) = value.errors.clone() {
+            io_error_count = errors.io_error_count;
+            io_error_threshold = errors.io_error_threshold;
+            io_stalled = errors.io_stalled;
+            io_stall_transition_count = errors.io_stall_transition_count;
+            io_stall_transition_threshold = errors.io_stall_transition_threshold;
+            if let Some(alerts) = errors.alerts {
+                alert_status = alerts.status;
+
+                notice = alerts.notice;
+
+                attention = alerts.attention;
+
+                warning = alerts.warning;
+
+                critical = alerts.critical;
+            }
+        }
         Self {
             name: value.name,
             used: value.used,
@@ -65,6 +153,16 @@ impl From<rpc::v1::pool::Pool> for PoolInfo {
             committed: value.committed,
             disk_capacity: value.disk_capacity,
             max_expandable_size: value.max_expandable_size.unwrap_or_default(),
+            io_error_count,
+            io_error_threshold,
+            io_stalled,
+            io_stall_transition_count,
+            io_stall_transition_threshold,
+            alert_status,
+            notice,
+            attention,
+            warning,
+            critical,
         }
     }
 }

metrics-exporter/src/bin/io_engine/collector/mod.rs

@@ -24,6 +24,43 @@ fn init_diskpool_gauge_vec(
     gauge_vec
 }
 
+/// Initializes a GaugeVec metric for diskpool alert reason with the provided metric name, description and
+/// descriptors.
+fn init_diskpool_alert_reason_gauge_vec(
+    metric_name: &str,
+    metric_desc: &str,
+    descs: &mut Vec<Desc>,
+) -> GaugeVec {
+    let opts = Opts::new(metric_name, metric_desc)
+        .subsystem("diskpool_alert")
+        .variable_labels(vec![
+            "node".to_string(),
+            "name".to_string(),
+            "unknown".to_string(),
+            "io_stalled".to_string(),
+            "io_stall_intermittent".to_string(),
+            "io_stall_intermittent_exc".to_string(),
+            "io_error".to_string(),
+            "io_error_exc".to_string(),
+        ]);
+    let gauge_vec = GaugeVec::new(
+        opts,
+        &[
+            "node",
+            "name",
+            "unknown",
+            "io_stalled",
+            "io_stall_intermittent",
+            "io_stall_intermittent_exc",
+            "io_error",
+            "io_error_exc",
+        ],
+    )
+    .unwrap_or_else(|_| panic!("Unable to create gauge metric type for {metric_name}"));
+    descs.extend(gauge_vec.desc().into_iter().cloned());
+    gauge_vec
+}
+
 /// Initializes a GaugeVec metric for volume with the provided metric name, description and
 /// descriptors.
 fn init_volume_gauge_vec(metric_name: &str, metric_desc: &str, descs: &mut Vec<Desc>) -> GaugeVec {