feat: new crate trace_metric for collecting metrics in read procedure (apache#714)

* feat: add ReadMetricsCollector for collecting metrics in read procedure

* collect metrics during merge

* add new component trace_metric

* add derive crate

* impl proc macro for trace_metric

* use the trace_metric crate

* support optional collector

* fix license

* make collector is optional

* trace metrics in parquet reader

* rename Collector to MetricsCollector

* rename TracedMetrics to TraceMetricWhenDrop

* support FormatCollectorVisitor

* set the metrics into datafusion

* remove useless file

* remove useless module

* rename constant variable

* fix wrong field name in metrics

* address CR

* attach index to the merge iter's metrics collector

* rename some constant variables

* fix wrong variable name

* rename metric type

Author: ShiKaiWi

Cargo.lock

@@ -855,6 +855,7 @@ impl SpaceStore {
             let sequence = table_data.last_sequence();
             let mut builder = MergeBuilder::new(MergeConfig {
                 request_id,
+                metrics_collector: None,
                 // no need to set deadline for compaction
                 deadline: None,
                 space_id,
@@ -1053,6 +1054,7 @@ fn build_mem_table_iter(memtable: MemTableRef, table_data: &TableData) -> Result
         projected_schema: ProjectedSchema::no_projection(table_data.schema()),
         need_dedup: table_data.dedup(),
         reverse: false,
+        metrics_collector: None,
     };
     memtable
         .scan(scan_ctx, scan_req)

analytic_engine/src/instance/flush_compaction.rs

@@ -23,6 +23,7 @@ use table_engine::{
     table::ReadRequest,
 };
 use tokio::sync::mpsc::{self, Receiver};
+use trace_metric::Metric;
 
 use crate::{
     instance::Instance,
@@ -66,6 +67,9 @@ pub enum Error {
 define_result!(Error);
 
 const RECORD_BATCH_READ_BUF_SIZE: usize = 1000;
+const MERGE_SORT_METRIC_NAME: &str = "do_merge_sort";
+const ITER_NUM_METRIC_NAME: &str = "iter_num";
+const MERGE_ITER_METRICS_COLLECTOR_NAME_PREFIX: &str = "merge_iter";
 
 /// Check whether it needs to apply merge sorting when reading the table with
 /// the `table_options` by the `read_request`.
@@ -91,13 +95,18 @@ impl Instance {
 
         let table_data = space_table.table_data();
 
-        // Collect metrics.
-        table_data.metrics.on_read_request_begin();
-
         let iter_options = self.iter_options.clone();
         let table_options = table_data.table_options();
 
-        if need_merge_sort_streams(&table_data.table_options(), &request) {
+        // Collect metrics.
+        table_data.metrics.on_read_request_begin();
+        let need_merge_sort = need_merge_sort_streams(&table_options, &request);
+        request.metrics_collector.collect(Metric::boolean(
+            MERGE_SORT_METRIC_NAME.to_string(),
+            need_merge_sort,
+        ));
+
+        if need_merge_sort {
             let merge_iters = self
                 .build_merge_iters(table_data, &request, iter_options, &table_options)
                 .await?;
@@ -123,16 +132,16 @@ impl Instance {
         };
 
         // Split iterators into `read_parallelism` groups.
-        let mut splited_iters: Vec<_> = std::iter::repeat_with(Vec::new)
+        let mut splitted_iters: Vec<_> = std::iter::repeat_with(Vec::new)
             .take(read_parallelism)
             .collect();
 
         for (i, time_aligned_iter) in partitioned_iters.into_iter().enumerate() {
-            splited_iters[i % read_parallelism].push(time_aligned_iter);
+            splitted_iters[i % read_parallelism].push(time_aligned_iter);
         }
 
         let mut streams = Vec::with_capacity(read_parallelism);
-        for iters in splited_iters {
+        for iters in splitted_iters {
             let stream = iters_to_stream(iters, self.read_runtime(), &request.projected_schema);
             streams.push(stream);
         }
@@ -169,9 +178,13 @@ impl Instance {
         let read_views = self.partition_ssts_and_memtables(time_range, version, table_options);
 
         let mut iters = Vec::with_capacity(read_views.len());
-        for read_view in read_views {
+        for (idx, read_view) in read_views.into_iter().enumerate() {
+            let metrics_collector = request
+                .metrics_collector
+                .span(format!("{MERGE_ITER_METRICS_COLLECTOR_NAME_PREFIX}_{idx}"));
             let merge_config = MergeConfig {
                 request_id: request.request_id,
+                metrics_collector: Some(metrics_collector),
                 deadline: request.opts.deadline,
                 space_id: table_data.space_id,
                 table_id: table_data.id,
@@ -201,6 +214,11 @@ impl Instance {
             iters.push(dedup_iter);
         }
 
+        request.metrics_collector.collect(Metric::number(
+            ITER_NUM_METRIC_NAME.to_string(),
+            iters.len(),
+        ));
+
         Ok(iters)
     }
 

analytic_engine/src/instance/read.rs

@@ -18,6 +18,7 @@ use common_types::{
 };
 use common_util::{define_result, error::GenericError};
 use snafu::{Backtrace, Snafu};
+use trace_metric::MetricsCollector;
 
 use crate::memtable::key::KeySequence;
 
@@ -129,14 +130,16 @@ pub struct ScanRequest {
     pub projected_schema: ProjectedSchema,
     pub need_dedup: bool,
     pub reverse: bool,
+    /// Collector for scan metrics.
+    pub metrics_collector: Option<MetricsCollector>,
 }
 
 /// In memory storage for table's data.
 ///
 /// # Concurrency
-/// The memtable is designed for single-writer and mutltiple-reader usage, so
+/// The memtable is designed for single-writer and multiple-reader usage, so
 /// not all function supports concurrent writer, the caller should guarantee not
-/// writing to the memtable concurrrently.
+/// writing to the memtable concurrently.
 // All operation is done in memory, no need to use async trait
 pub trait MemTable {
     /// Schema of this memtable

analytic_engine/src/memtable/mod.rs

@@ -204,6 +204,7 @@ mod tests {
                     projected_schema: projected_schema.clone(),
                     need_dedup: true,
                     reverse: false,
+                    metrics_collector: None,
                 },
                 vec![
                     build_row(b"a", 1, 10.0, "v1", 1000, 1_000_000),
@@ -223,6 +224,7 @@ mod tests {
                     projected_schema: projected_schema.clone(),
                     need_dedup: true,
                     reverse: false,
+                    metrics_collector: None,
                 },
                 vec![
                     build_row(b"a", 1, 10.0, "v1", 1000, 1_000_000),
@@ -241,6 +243,7 @@ mod tests {
                     projected_schema,
                     need_dedup: true,
                     reverse: false,
+                    metrics_collector: None,
                 },
                 vec![
                     build_row(b"a", 1, 10.0, "v1", 1000, 1_000_000),
@@ -272,6 +275,7 @@ mod tests {
                 projected_schema,
                 need_dedup: true,
                 reverse: false,
+                metrics_collector: None,
             },
             vec![
                 build_row_for_two_column(b"a", 1),

analytic_engine/src/memtable/skiplist/mod.rs

@@ -116,6 +116,7 @@ impl<'a> Builder<'a> {
                 false,
                 self.config.predicate.as_ref(),
                 self.config.deadline,
+                None,
             )
             .context(BuildStreamFromMemtable)?;
             streams.push(stream);
@@ -131,6 +132,7 @@ impl<'a> Builder<'a> {
                 false,
                 self.config.predicate.as_ref(),
                 self.config.deadline,
+                None,
             )
             .context(BuildStreamFromMemtable)?;
             streams.push(stream);
@@ -145,6 +147,7 @@ impl<'a> Builder<'a> {
                     self.config.sst_factory,
                     &self.config.sst_read_options,
                     self.config.store_picker,
+                    None,
                 )
                 .await
                 .context(BuildStreamFromSst)?;