TMP WIP

disk/src/binned.rs

@@ -23,13 +23,14 @@ use futures_util::{FutureExt, StreamExt};
 use netpod::log::*;
 use netpod::timeunits::SEC;
 use netpod::{
-    BinnedRange, ByteOrder, NanoRange, NodeConfigCached, PerfOpts, PreBinnedPatchIterator, PreBinnedPatchRange,
-    ScalarType, Shape,
+    AggKind, BinnedRange, ByteOrder, NanoRange, NodeConfigCached, PerfOpts, PreBinnedPatchIterator,
+    PreBinnedPatchRange, ScalarType, Shape,
 };
-use num_traits::{AsPrimitive, Bounded, Zero};
+use num_traits::{AsPrimitive, Bounded, Float, Zero};
 use parse::channelconfig::{extract_matching_config_entry, read_local_config, MatchingConfigEntry};
 use serde::de::DeserializeOwned;
 use serde::{Deserialize, Serialize, Serializer};
+use std::fmt;
 use std::fmt::Debug;
 use std::future::Future;
 use std::marker::PhantomData;
@@ -122,8 +123,13 @@ where
                 range: query.range().clone(),
                 agg_kind: query.agg_kind().clone(),
             };
+            let x_bin_count = if let AggKind::DimXBinsN(n) = query.agg_kind() {
+                *n as usize
+            } else {
+                0
+            };
             let s = MergedFromRemotes::<ENP>::new(evq, perf_opts, node_config.node_config.cluster.clone());
-            let s = TBinnerStream::<_, <ENP as EventsNodeProcessor>::Output>::new(s, range);
+            let s = TBinnerStream::<_, <ENP as EventsNodeProcessor>::Output>::new(s, range, x_bin_count);
             let ret = BinnedResponseStat {
                 stream: Box::pin(s),
                 bin_count,
@@ -770,13 +776,47 @@ pub trait NumOps:
     + Serialize
     + DeserializeOwned
 {
+    fn min_or_nan() -> Self;
+    fn max_or_nan() -> Self;
+    fn is_nan(&self) -> bool;
 }
 
-impl<T> NumOps for T where
-    T: Send + Unpin + Debug + Zero + AsPrimitive<f32> + Bounded + PartialOrd + SubFrId + Serialize + DeserializeOwned
-{
+fn tmp() {}
+macro_rules! impl_num_ops {
+    ($ty:ident, $min_or_nan:ident, $max_or_nan:ident, $is_nan:ident) => {
+        impl NumOps for $ty {
+            fn min_or_nan() -> Self {
+                $ty::$min_or_nan
+            }
+            fn max_or_nan() -> Self {
+                $ty::$max_or_nan
+            }
+            fn is_nan(&self) -> bool {
+                $is_nan(self)
+            }
+        }
+    };
 }
 
+fn is_nan_int<T>(x: &T) -> bool {
+    false
+}
+
+fn is_nan_float<T: Float>(x: &T) -> bool {
+    x.is_nan()
+}
+
+impl_num_ops!(u8, MIN, MAX, is_nan_int);
+impl_num_ops!(u16, MIN, MAX, is_nan_int);
+impl_num_ops!(u32, MIN, MAX, is_nan_int);
+impl_num_ops!(u64, MIN, MAX, is_nan_int);
+impl_num_ops!(i8, MIN, MAX, is_nan_int);
+impl_num_ops!(i16, MIN, MAX, is_nan_int);
+impl_num_ops!(i32, MIN, MAX, is_nan_int);
+impl_num_ops!(i64, MIN, MAX, is_nan_int);
+impl_num_ops!(f32, NAN, NAN, is_nan_float);
+impl_num_ops!(f64, NAN, NAN, is_nan_float);
+
 pub trait EventsDecoder {
     type Output;
     fn ingest(&mut self, event: &[u8]);
@@ -786,7 +826,8 @@ pub trait EventsDecoder {
 pub trait EventsNodeProcessor: Send + Unpin {
     type Input;
     type Output: Send + Unpin + DeserializeOwned + WithTimestamps + TimeBinnableType;
-    fn process(inp: EventValues<Self::Input>) -> Self::Output;
+    fn create(shape: Shape) -> Self;
+    fn process(&self, inp: EventValues<Self::Input>) -> Self::Output;
 }
 
 pub trait TimeBins: Send + Unpin + WithLen + Appendable + FilterFittingInside {
@@ -799,16 +840,17 @@ pub struct MinMaxAvgBins<NTY> {
     pub ts1s: Vec<u64>,
     pub ts2s: Vec<u64>,
     pub counts: Vec<u64>,
+    // TODO get rid of Option:
     pub mins: Vec<Option<NTY>>,
     pub maxs: Vec<Option<NTY>>,
     pub avgs: Vec<Option<f32>>,
 }
 
-impl<NTY> std::fmt::Debug for MinMaxAvgBins<NTY>
+impl<NTY> fmt::Debug for MinMaxAvgBins<NTY>
 where
-    NTY: std::fmt::Debug,
+    NTY: fmt::Debug,
 {
-    fn fmt(&self, fmt: &mut std::fmt::Formatter) -> std::fmt::Result {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result {
         write!(
             fmt,
             "MinMaxAvgBins  count {}  ts1s {:?}  ts2s {:?}  counts {:?}  mins {:?}  maxs {:?}  avgs {:?}",
@@ -951,7 +993,7 @@ where
     type Output = MinMaxAvgBins<NTY>;
     type Aggregator = MinMaxAvgBinsAggregator<NTY>;
 
-    fn aggregator(range: NanoRange) -> Self::Aggregator {
+    fn aggregator(range: NanoRange, bin_count: usize) -> Self::Aggregator {
         Self::Aggregator::new(range)
     }
 }
@@ -1103,7 +1145,8 @@ impl<NTY> EventValuesAggregator<NTY> {
         Self {
             range,
             count: 0,
-            min: None,
+            // TODO get rid of Option
+            min: err::todoval(),
             max: None,
             sumc: 0,
             sum: 0f32,
@@ -1282,3 +1325,400 @@ pub enum RangeCompletableItem<T> {
     RangeComplete,
     Data(T),
 }
+
+#[derive(Clone, Serialize, Deserialize)]
+pub struct MinMaxAvgWaveBins<NTY> {
+    pub ts1s: Vec<u64>,
+    pub ts2s: Vec<u64>,
+    pub counts: Vec<u64>,
+    pub mins: Vec<Option<Vec<NTY>>>,
+    pub maxs: Vec<Option<Vec<NTY>>>,
+    pub avgs: Vec<Option<Vec<f32>>>,
+}
+
+impl<NTY> fmt::Debug for MinMaxAvgWaveBins<NTY>
+where
+    NTY: fmt::Debug,
+{
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result {
+        write!(
+            fmt,
+            "MinMaxAvgBins  count {}  ts1s {:?}  ts2s {:?}  counts {:?}  mins {:?}  maxs {:?}  avgs {:?}",
+            self.ts1s.len(),
+            self.ts1s.iter().map(|k| k / SEC).collect::<Vec<_>>(),
+            self.ts2s.iter().map(|k| k / SEC).collect::<Vec<_>>(),
+            self.counts,
+            self.mins,
+            self.maxs,
+            self.avgs,
+        )
+    }
+}
+
+impl<NTY> MinMaxAvgWaveBins<NTY> {
+    pub fn empty() -> Self {
+        Self {
+            ts1s: vec![],
+            ts2s: vec![],
+            counts: vec![],
+            mins: vec![],
+            maxs: vec![],
+            avgs: vec![],
+        }
+    }
+}
+
+impl<NTY> FitsInside for MinMaxAvgWaveBins<NTY> {
+    fn fits_inside(&self, range: NanoRange) -> Fits {
+        if self.ts1s.is_empty() {
+            Fits::Empty
+        } else {
+            let t1 = *self.ts1s.first().unwrap();
+            let t2 = *self.ts2s.last().unwrap();
+            if t2 <= range.beg {
+                Fits::Lower
+            } else if t1 >= range.end {
+                Fits::Greater
+            } else if t1 < range.beg && t2 > range.end {
+                Fits::PartlyLowerAndGreater
+            } else if t1 < range.beg {
+                Fits::PartlyLower
+            } else if t2 > range.end {
+                Fits::PartlyGreater
+            } else {
+                Fits::Inside
+            }
+        }
+    }
+}
+
+impl<NTY> FilterFittingInside for MinMaxAvgWaveBins<NTY> {
+    fn filter_fitting_inside(self, fit_range: NanoRange) -> Option<Self> {
+        match self.fits_inside(fit_range) {
+            Fits::Inside | Fits::PartlyGreater | Fits::PartlyLower | Fits::PartlyLowerAndGreater => Some(self),
+            _ => None,
+        }
+    }
+}
+
+impl<NTY> RangeOverlapInfo for MinMaxAvgWaveBins<NTY> {
+    fn ends_before(&self, range: NanoRange) -> bool {
+        match self.ts2s.last() {
+            Some(&ts) => ts <= range.beg,
+            None => true,
+        }
+    }
+
+    fn ends_after(&self, range: NanoRange) -> bool {
+        match self.ts2s.last() {
+            Some(&ts) => ts > range.end,
+            None => panic!(),
+        }
+    }
+
+    fn starts_after(&self, range: NanoRange) -> bool {
+        match self.ts1s.first() {
+            Some(&ts) => ts >= range.end,
+            None => panic!(),
+        }
+    }
+}
+
+impl<NTY> TimeBins for MinMaxAvgWaveBins<NTY>
+where
+    NTY: NumOps,
+{
+    fn ts1s(&self) -> &Vec<u64> {
+        &self.ts1s
+    }
+
+    fn ts2s(&self) -> &Vec<u64> {
+        &self.ts2s
+    }
+}
+
+impl<NTY> WithLen for MinMaxAvgWaveBins<NTY> {
+    fn len(&self) -> usize {
+        self.ts1s.len()
+    }
+}
+
+impl<NTY> Appendable for MinMaxAvgWaveBins<NTY>
+where
+    NTY: NumOps,
+{
+    fn empty() -> Self {
+        Self::empty()
+    }
+
+    fn append(&mut self, src: &Self) {
+        self.ts1s.extend_from_slice(&src.ts1s);
+        self.ts2s.extend_from_slice(&src.ts2s);
+        self.counts.extend_from_slice(&src.counts);
+        self.mins.extend_from_slice(&src.mins);
+        self.maxs.extend_from_slice(&src.maxs);
+        self.avgs.extend_from_slice(&src.avgs);
+    }
+}
+
+impl<NTY> ReadableFromFile for MinMaxAvgWaveBins<NTY>
+where
+    NTY: NumOps,
+{
+    // TODO this function is not needed in the trait:
+    fn read_from_file(file: File) -> Result<ReadPbv<Self>, Error> {
+        Ok(ReadPbv::new(file))
+    }
+
+    fn from_buf(buf: &[u8]) -> Result<Self, Error> {
+        let dec = serde_cbor::from_slice(&buf)?;
+        Ok(dec)
+    }
+}
+
+impl<NTY> TimeBinnableType for MinMaxAvgWaveBins<NTY>
+where
+    NTY: NumOps,
+{
+    type Output = MinMaxAvgWaveBins<NTY>;
+    type Aggregator = MinMaxAvgWaveBinsAggregator<NTY>;
+
+    fn aggregator(range: NanoRange, x_bin_count: usize) -> Self::Aggregator {
+        Self::Aggregator::new(range, x_bin_count)
+    }
+}
+
+impl<NTY> ToJsonResult for Sitemty<MinMaxAvgWaveBins<NTY>>
+where
+    NTY: NumOps,
+{
+    fn to_json_result(&self) -> Result<Box<dyn ToJsonBytes>, Error> {
+        Ok(Box::new(serde_json::Value::String(format!(
+            "MinMaxAvgBins/non-json-item"
+        ))))
+    }
+}
+
+pub struct MinMaxAvgWaveBinsCollected<NTY> {
+    _m1: PhantomData<NTY>,
+}
+
+impl<NTY> MinMaxAvgWaveBinsCollected<NTY> {
+    pub fn new() -> Self {
+        Self { _m1: PhantomData }
+    }
+}
+
+#[derive(Serialize)]
+pub struct MinMaxAvgWaveBinsCollectedResult<NTY> {
+    ts0: u64,
+    tsoff: Vec<u64>,
+    //ts_bin_edges: Vec<IsoDateTime>,
+    counts: Vec<u64>,
+    mins: Vec<Option<Vec<NTY>>>,
+    maxs: Vec<Option<Vec<NTY>>>,
+    avgs: Vec<Option<Vec<f32>>>,
+    #[serde(skip_serializing_if = "Bool::is_false", rename = "finalisedRange")]
+    finalised_range: bool,
+    #[serde(skip_serializing_if = "Zero::is_zero", rename = "missingBins")]
+    missing_bins: u32,
+    #[serde(skip_serializing_if = "Option::is_none", rename = "continueAt")]
+    //continue_at: Option<IsoDateTime>,
+    continue_at: Option<u64>,
+}
+
+pub struct MinMaxAvgWaveBinsCollector<NTY> {
+    bin_count_exp: u32,
+    timed_out: bool,
+    range_complete: bool,
+    vals: MinMaxAvgWaveBins<NTY>,
+    _m1: PhantomData<NTY>,
+}
+
+impl<NTY> MinMaxAvgWaveBinsCollector<NTY> {
+    pub fn new(bin_count_exp: u32) -> Self {
+        Self {
+            bin_count_exp,
+            timed_out: false,
+            range_complete: false,
+            vals: MinMaxAvgWaveBins::<NTY>::empty(),
+            _m1: PhantomData,
+        }
+    }
+}
+
+impl<NTY> WithLen for MinMaxAvgWaveBinsCollector<NTY>
+where
+    NTY: NumOps + Serialize,
+{
+    fn len(&self) -> usize {
+        self.vals.ts1s.len()
+    }
+}
+
+impl<NTY> Collector for MinMaxAvgWaveBinsCollector<NTY>
+where
+    NTY: NumOps + Serialize,
+{
+    type Input = MinMaxAvgWaveBins<NTY>;
+    type Output = MinMaxAvgWaveBinsCollectedResult<NTY>;
+
+    fn ingest(&mut self, src: &Self::Input) {
+        Appendable::append(&mut self.vals, src);
+    }
+
+    fn set_range_complete(&mut self) {
+        self.range_complete = true;
+    }
+
+    fn set_timed_out(&mut self) {
+        self.timed_out = true;
+    }
+
+    fn result(self) -> Result<Self::Output, Error> {
+        let ts0 = self.vals.ts1s.first().map_or(0, |k| *k / SEC);
+        let bin_count = self.vals.ts1s.len() as u32;
+        let mut tsoff: Vec<_> = self.vals.ts1s.iter().map(|k| *k - ts0 * SEC).collect();
+        if let Some(&k) = self.vals.ts2s.last() {
+            tsoff.push(k - ts0 * SEC);
+        }
+        let tsoff = tsoff;
+        let _iso: Vec<_> = tsoff
+            .iter()
+            .map(|&k| IsoDateTime(Utc.timestamp_nanos(k as i64)))
+            .collect();
+        let continue_at = if self.vals.ts1s.len() < self.bin_count_exp as usize {
+            match tsoff.last() {
+                Some(k) => Some(k.clone()),
+                None => Err(Error::with_msg("partial_content but no bin in result"))?,
+            }
+        } else {
+            None
+        };
+        let ret = MinMaxAvgWaveBinsCollectedResult {
+            ts0,
+            tsoff,
+            counts: self.vals.counts,
+            mins: self.vals.mins,
+            maxs: self.vals.maxs,
+            avgs: self.vals.avgs,
+            finalised_range: self.range_complete,
+            missing_bins: self.bin_count_exp - bin_count,
+            continue_at,
+        };
+        Ok(ret)
+    }
+}
+
+impl<NTY> Collectable for MinMaxAvgWaveBins<NTY>
+where
+    NTY: NumOps + Serialize,
+{
+    type Collector = MinMaxAvgWaveBinsCollector<NTY>;
+
+    fn new_collector(bin_count_exp: u32) -> Self::Collector {
+        Self::Collector::new(bin_count_exp)
+    }
+}
+
+pub struct MinMaxAvgWaveBinsAggregator<NTY> {
+    range: NanoRange,
+    count: u64,
+    min: Vec<NTY>,
+    max: Vec<NTY>,
+    sum: Vec<f32>,
+    sumc: u64,
+}
+
+impl<NTY> MinMaxAvgWaveBinsAggregator<NTY>
+where
+    NTY: NumOps,
+{
+    pub fn new(range: NanoRange, x_bin_count: usize) -> Self {
+        Self {
+            range,
+            count: 0,
+            min: vec![NTY::min_or_nan(); x_bin_count],
+            max: vec![NTY::max_or_nan(); x_bin_count],
+            sum: vec![0f32; x_bin_count],
+            sumc: 0,
+        }
+    }
+}
+
+impl<NTY> TimeBinnableTypeAggregator for MinMaxAvgWaveBinsAggregator<NTY>
+where
+    NTY: NumOps,
+{
+    type Input = MinMaxAvgWaveBins<NTY>;
+    type Output = MinMaxAvgWaveBins<NTY>;
+
+    fn range(&self) -> &NanoRange {
+        &self.range
+    }
+
+    fn ingest(&mut self, item: &Self::Input) {
+        for i1 in 0..item.ts1s.len() {
+            if item.ts2s[i1] <= self.range.beg {
+                continue;
+            } else if item.ts1s[i1] >= self.range.end {
+                continue;
+            } else {
+                // the input can contain bins where no events did fall into.
+                match &item.mins[i1] {
+                    None => {}
+                    Some(inp) => {
+                        for (a, b) in self.min.iter_mut().zip(inp.iter()) {
+                            if *b < *a {
+                                *a = *b;
+                            }
+                        }
+                    }
+                }
+                match &item.maxs[i1] {
+                    None => {}
+                    Some(inp) => {
+                        for (a, b) in self.max.iter_mut().zip(inp.iter()) {
+                            if *b > *a {
+                                *a = *b;
+                            }
+                        }
+                    }
+                }
+                match &item.avgs[i1] {
+                    None => {}
+                    Some(inp) => {
+                        for (a, b) in self.sum.iter_mut().zip(inp.iter()) {
+                            *a += *b;
+                        }
+                    }
+                }
+                self.sumc += 1;
+                self.count += item.counts[i1];
+            }
+        }
+    }
+
+    fn result(self) -> Self::Output {
+        if self.sumc == 0 {
+            Self::Output {
+                ts1s: vec![self.range.beg],
+                ts2s: vec![self.range.end],
+                counts: vec![self.count],
+                mins: vec![None],
+                maxs: vec![None],
+                avgs: vec![None],
+            }
+        } else {
+            let avg = self.sum.iter().map(|j| *j / self.sumc as f32).collect();
+            Self::Output {
+                ts1s: vec![self.range.beg],
+                ts2s: vec![self.range.end],
+                counts: vec![self.count],
+                mins: vec![Some(self.min)],
+                maxs: vec![Some(self.max)],
+                avgs: vec![Some(avg)],
+            }
+        }
+    }
+}