feat(pageserver): add metrics for aux file size (#7623)

ref #7443

## Summary of changes

This pull request adds a size estimator for aux files. Each timeline
stores a cached `isize` for the estimated total size of aux files. It
gets reset on basebackup, and gets updated for each aux file
modification. TODO: print a warning when it exceeds the size.

The size metrics is not accurate. Race between `on_basebackup` and other
functions could create a negative basebackup size, but the chance is
rare. Anyways, this does not impose any extra I/Os to the storage as
everything is computed in-memory.

The aux files are only stored on shard 0. As basebackups are only
generated on shard 0, only shard 0 will report this metrics.

---------

Signed-off-by: Alex Chi Z <[email protected]>

pageserver/src/aux_file.rs

@@ -1,3 +1,6 @@
+use std::sync::Arc;
+
+use ::metrics::IntGauge;
 use bytes::{Buf, BufMut, Bytes};
 use pageserver_api::key::{Key, AUX_KEY_PREFIX, METADATA_KEY_SIZE};
 use tracing::warn;
@@ -140,6 +143,55 @@ pub fn encode_file_value(files: &[(&str, &[u8])]) -> anyhow::Result<Vec<u8>> {
     Ok(encoded)
 }
 
+/// An estimation of the size of aux files.
+pub struct AuxFileSizeEstimator {
+    aux_file_size_gauge: IntGauge,
+    size: Arc<std::sync::Mutex<Option<isize>>>,
+}
+
+impl AuxFileSizeEstimator {
+    pub fn new(aux_file_size_gauge: IntGauge) -> Self {
+        Self {
+            aux_file_size_gauge,
+            size: Arc::new(std::sync::Mutex::new(None)),
+        }
+    }
+
+    pub fn on_base_backup(&self, new_size: usize) {
+        let mut guard = self.size.lock().unwrap();
+        *guard = Some(new_size as isize);
+        self.report(new_size as isize);
+    }
+
+    pub fn on_add(&self, file_size: usize) {
+        let mut guard = self.size.lock().unwrap();
+        if let Some(size) = &mut *guard {
+            *size += file_size as isize;
+            self.report(*size);
+        }
+    }
+
+    pub fn on_remove(&self, file_size: usize) {
+        let mut guard = self.size.lock().unwrap();
+        if let Some(size) = &mut *guard {
+            *size -= file_size as isize;
+            self.report(*size);
+        }
+    }
+
+    pub fn on_update(&self, old_size: usize, new_size: usize) {
+        let mut guard = self.size.lock().unwrap();
+        if let Some(size) = &mut *guard {
+            *size += new_size as isize - old_size as isize;
+            self.report(*size);
+        }
+    }
+
+    pub fn report(&self, size: isize) {
+        self.aux_file_size_gauge.set(size as i64);
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

pageserver/src/metrics.rs

@@ -585,6 +585,15 @@ static CURRENT_LOGICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     .expect("failed to define current logical size metric")
 });
 
+static AUX_FILE_SIZE: Lazy<IntGaugeVec> = Lazy::new(|| {
+    register_int_gauge_vec!(
+        "pageserver_aux_file_estimated_size",
+        "The size of all aux files for a timeline in aux file v2 store.",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) mod initial_logical_size {
     use metrics::{register_int_counter, register_int_counter_vec, IntCounter, IntCounterVec};
     use once_cell::sync::Lazy;
@@ -2115,6 +2124,7 @@ pub(crate) struct TimelineMetrics {
     resident_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
     pub current_logical_size_gauge: UIntGauge,
+    pub aux_file_size_gauge: IntGauge,
     pub directory_entries_count_gauge: Lazy<UIntGauge, Box<dyn Send + Fn() -> UIntGauge>>,
     pub evictions: IntCounter,
     pub evictions_with_low_residence_duration: std::sync::RwLock<EvictionsWithLowResidenceDuration>,
@@ -2187,6 +2197,9 @@ impl TimelineMetrics {
         let current_logical_size_gauge = CURRENT_LOGICAL_SIZE
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
+        let aux_file_size_gauge = AUX_FILE_SIZE
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
         // TODO use impl Trait syntax here once we have ability to use it: https://github.com/rust-lang/rust/issues/63065
         let directory_entries_count_gauge_closure = {
             let tenant_shard_id = *tenant_shard_id;
@@ -2224,6 +2237,7 @@ impl TimelineMetrics {
             last_record_gauge,
             resident_physical_size_gauge,
             current_logical_size_gauge,
+            aux_file_size_gauge,
             directory_entries_count_gauge,
             evictions,
             evictions_with_low_residence_duration: std::sync::RwLock::new(
@@ -2264,6 +2278,7 @@ impl TimelineMetrics {
             let _ = metric.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         }
         let _ = EVICTIONS.remove_label_values(&[tenant_id, shard_id, timeline_id]);
+        let _ = AUX_FILE_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
 
         self.evictions_with_low_residence_duration
             .write()

pageserver/src/pgdatadir_mapping.rs

@@ -699,13 +699,17 @@ impl Timeline {
             .await
             .context("scan")?;
         let mut result = HashMap::new();
+        let mut sz = 0;
         for (_, v) in kv {
             let v = v.context("get value")?;
             let v = aux_file::decode_file_value_bytes(&v).context("value decode")?;
             for (fname, content) in v {
+                sz += fname.len();
+                sz += content.len();
                 result.insert(fname, content);
             }
         }
+        self.aux_file_size_estimator.on_base_backup(sz);
         Ok(result)
     }
 
@@ -1474,23 +1478,45 @@ impl<'a> DatadirModification<'a> {
                 Err(PageReconstructError::MissingKey(_)) => None,
                 Err(e) => return Err(e.into()),
             };
-            let files = if let Some(ref old_val) = old_val {
+            let files: Vec<(&str, &[u8])> = if let Some(ref old_val) = old_val {
                 aux_file::decode_file_value(old_val)?
             } else {
                 Vec::new()
             };
-            let new_files = if content.is_empty() {
-                files
-                    .into_iter()
-                    .filter(|(p, _)| &path != p)
-                    .collect::<Vec<_>>()
-            } else {
-                files
-                    .into_iter()
-                    .filter(|(p, _)| &path != p)
-                    .chain(std::iter::once((path, content)))
-                    .collect::<Vec<_>>()
-            };
+            let mut other_files = Vec::with_capacity(files.len());
+            let mut modifying_file = None;
+            for file @ (p, content) in files {
+                if path == p {
+                    assert!(
+                        modifying_file.is_none(),
+                        "duplicated entries found for {}",
+                        path
+                    );
+                    modifying_file = Some(content);
+                } else {
+                    other_files.push(file);
+                }
+            }
+            let mut new_files = other_files;
+            match (modifying_file, content.is_empty()) {
+                (Some(old_content), false) => {
+                    self.tline
+                        .aux_file_size_estimator
+                        .on_update(old_content.len(), content.len());
+                    new_files.push((path, content));
+                }
+                (Some(old_content), true) => {
+                    self.tline
+                        .aux_file_size_estimator
+                        .on_remove(old_content.len());
+                    // not adding the file key to the final `new_files` vec.
+                }
+                (None, false) => {
+                    self.tline.aux_file_size_estimator.on_add(content.len());
+                    new_files.push((path, content));
+                }
+                (None, true) => anyhow::bail!("removing non-existing aux file: {}", path),
+            }
             let new_val = aux_file::encode_file_value(&new_files)?;
             self.put(key, Value::Image(new_val.into()));
         }

pageserver/src/tenant/timeline.rs

@@ -61,9 +61,12 @@ use std::{
 };
 
 use crate::tenant::timeline::init::LocalLayerFileMetadata;
-use crate::tenant::{
-    layer_map::{LayerMap, SearchResult},
-    metadata::TimelineMetadata,
+use crate::{
+    aux_file::AuxFileSizeEstimator,
+    tenant::{
+        layer_map::{LayerMap, SearchResult},
+        metadata::TimelineMetadata,
+    },
 };
 use crate::{
     context::{DownloadBehavior, RequestContext},
@@ -409,6 +412,8 @@ pub struct Timeline {
 
     /// Keep aux directory cache to avoid it's reconstruction on each update
     pub(crate) aux_files: tokio::sync::Mutex<AuxFilesState>,
+
+    pub(crate) aux_file_size_estimator: AuxFileSizeEstimator,
 }
 
 pub struct WalReceiverInfo {
@@ -2161,6 +2166,16 @@ impl Timeline {
         };
 
         Arc::new_cyclic(|myself| {
+            let metrics = TimelineMetrics::new(
+                &tenant_shard_id,
+                &timeline_id,
+                crate::metrics::EvictionsWithLowResidenceDurationBuilder::new(
+                    "mtime",
+                    evictions_low_residence_duration_metric_threshold,
+                ),
+            );
+            let aux_file_metrics = metrics.aux_file_size_gauge.clone();
+
             let mut result = Timeline {
                 conf,
                 tenant_conf,
@@ -2192,14 +2207,7 @@ impl Timeline {
                 ancestor_timeline: ancestor,
                 ancestor_lsn: metadata.ancestor_lsn(),
 
-                metrics: TimelineMetrics::new(
-                    &tenant_shard_id,
-                    &timeline_id,
-                    crate::metrics::EvictionsWithLowResidenceDurationBuilder::new(
-                        "mtime",
-                        evictions_low_residence_duration_metric_threshold,
-                    ),
-                ),
+                metrics,
 
                 query_metrics: crate::metrics::SmgrQueryTimePerTimeline::new(
                     &tenant_shard_id,
@@ -2263,6 +2271,8 @@ impl Timeline {
                     dir: None,
                     n_deltas: 0,
                 }),
+
+                aux_file_size_estimator: AuxFileSizeEstimator::new(aux_file_metrics),
             };
             result.repartition_threshold =
                 result.get_checkpoint_distance() / REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE;

test_runner/fixtures/metrics.py

@@ -149,6 +149,7 @@ def histogram(prefix_without_trailing_underscore: str) -> List[str]:
     "pageserver_storage_operations_seconds_sum_total",
     "pageserver_evictions_total",
     "pageserver_evictions_with_low_residence_duration_total",
+    "pageserver_aux_file_estimated_size",
     *PAGESERVER_PER_TENANT_REMOTE_TIMELINE_CLIENT_METRICS,
     # "pageserver_directory_entries_count", -- only used if above a certain threshold
     # "pageserver_broken_tenants_count" -- used only for broken