Remove temp file (#2815)

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>

src/directory/mmap_directory/mod.rs

@@ -676,7 +676,7 @@ mod tests {
             let num_segments = reader.searcher().segment_readers().len();
             assert!(num_segments <= 4);
             let num_components_except_deletes_and_tempstore =
-                crate::index::SegmentComponent::iterator().len() - 2;
+                crate::index::SegmentComponent::iterator().len() - 1;
             let max_num_mmapped = num_components_except_deletes_and_tempstore * num_segments;
             assert_eventually(|| {
                 let num_mmapped = mmap_directory.get_cache_info().mmapped.len();

src/docset.rs

@@ -51,31 +51,55 @@ pub trait DocSet: Send {
         doc
     }
 
-    /// Seeks to the target if possible and returns true if the target is in the DocSet.
+    /// !!!Dragons ahead!!!
+    /// In spirit, this is an approximate and dangerous version of `seek`.
+    ///
+    /// It can leave the DocSet in an `invalid` state and might return a
+    /// lower bound of what the result of Seek would have been.
+    ///
+    ///
+    /// More accurately it returns either:
+    /// - Found if the target is in the docset. In that case, the DocSet is left in a valid state.
+    /// - SeekLowerBound(seek_lower_bound) if the target is not in the docset. In that case, The
+    ///   DocSet can be the left in a invalid state. The DocSet should then only receives call to
+    ///   `seek_danger(..)` until it returns `Found`, and get back to a valid state.
+    ///
+    /// `seek_lower_bound` can be any `DocId` (in the docset or not) as long as it is in
+    /// `(target .. seek_result] U {TERMINATED}` where `seek_result` is the first document in the
+    /// docset greater than to `target`.
+    ///
+    /// `seek_danger` may return `SeekLowerBound(TERMINATED)`.
+    ///
+    /// Calling `seek_danger` with TERMINATED as a target is allowed,
+    /// and should always return NewTarget(TERMINATED) or anything larger as TERMINATED is NOT in
+    /// the DocSet.
     ///
     /// DocSets that already have an efficient `seek` method don't need to implement
-    /// `seek_into_the_danger_zone`. All wrapper DocSets should forward
-    /// `seek_into_the_danger_zone` to the underlying DocSet.
+    /// `seek_danger`.
     ///
-    /// ## API Behaviour
-    /// If `seek_into_the_danger_zone` is returning true, a call to `doc()` has to return target.
-    /// If `seek_into_the_danger_zone` is returning false, a call to `doc()` may return any doc
-    /// between the last doc that matched and target or a doc that is a valid next hit after
-    /// target. The DocSet is considered to be in an invalid state until
-    /// `seek_into_the_danger_zone` returns true again.
-    ///
-    /// `target` needs to be equal or larger than `doc` when in a valid state.
-    ///
-    /// Consecutive calls are not allowed to have decreasing `target` values.
-    ///
-    /// # Warning
-    /// This is an advanced API used by intersection. The API contract is tricky, avoid using it.
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        let current_doc = self.doc();
-        if current_doc < target {
-            self.seek(target);
+    /// Consecutive calls to seek_danger are guaranteed to have strictly increasing `target`
+    /// values.
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        if target >= TERMINATED {
+            debug_assert!(target == TERMINATED);
+            // No need to advance.
+            return SeekDangerResult::SeekLowerBound(target);
+        }
+
+        // The default implementation does not include any
+        // `danger zone` behavior.
+        //
+        // It does not leave the scorer in an invalid state.
+        // For this reason, we can safely call `self.doc()`.
+        let mut doc = self.doc();
+        if doc < target {
+            doc = self.seek(target);
+        }
+        if doc == target {
+            SeekDangerResult::Found
+        } else {
+            SeekDangerResult::SeekLowerBound(doc)
         }
-        self.doc() == target
     }
 
     /// Fills a given mutable buffer with the next doc ids from the
@@ -166,6 +190,17 @@ pub trait DocSet: Send {
     }
 }
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum SeekDangerResult {
+    /// The target was found in the DocSet.
+    Found,
+    /// The target was not found in the DocSet.
+    /// We return a range in which the value could be.
+    /// The given target can be any DocId, that is <= than the first document
+    /// in the docset after the target.
+    SeekLowerBound(DocId),
+}
+
 impl DocSet for &mut dyn DocSet {
     fn advance(&mut self) -> u32 {
         (**self).advance()
@@ -175,8 +210,8 @@ impl DocSet for &mut dyn DocSet {
         (**self).seek(target)
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        (**self).seek_into_the_danger_zone(target)
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        (**self).seek_danger(target)
     }
 
     fn doc(&self) -> u32 {
@@ -211,9 +246,9 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
         unboxed.seek(target)
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
         let unboxed: &mut TDocSet = self.borrow_mut();
-        unboxed.seek_into_the_danger_zone(target)
+        unboxed.seek_danger(target)
     }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId; COLLECT_BLOCK_BUFFER_LEN]) -> usize {

src/index/index_meta.rs

@@ -1,8 +1,6 @@
 use std::collections::HashSet;
 use std::fmt;
 use std::path::PathBuf;
-use std::sync::atomic::AtomicBool;
-use std::sync::Arc;
 
 use serde::{Deserialize, Serialize};
 
@@ -37,7 +35,6 @@ impl SegmentMetaInventory {
         let inner = InnerSegmentMeta {
             segment_id,
             max_doc,
-            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
             deletes: None,
         };
         SegmentMeta::from(self.inventory.track(inner))
@@ -85,15 +82,6 @@ impl SegmentMeta {
         self.tracked.segment_id
     }
 
-    /// Removes the Component::TempStore from the alive list and
-    /// therefore marks the temp docstore file to be deleted by
-    /// the garbage collection.
-    pub fn untrack_temp_docstore(&self) {
-        self.tracked
-            .include_temp_doc_store
-            .store(false, std::sync::atomic::Ordering::Relaxed);
-    }
-
     /// Returns the number of deleted documents.
     pub fn num_deleted_docs(&self) -> u32 {
         self.tracked
@@ -111,20 +99,9 @@ impl SegmentMeta {
     /// is by removing all files that have been created by tantivy
     /// and are not used by any segment anymore.
     pub fn list_files(&self) -> HashSet<PathBuf> {
-        if self
-            .tracked
-            .include_temp_doc_store
-            .load(std::sync::atomic::Ordering::Relaxed)
-        {
-            SegmentComponent::iterator()
-                .map(|component| self.relative_path(*component))
-                .collect::<HashSet<PathBuf>>()
-        } else {
-            SegmentComponent::iterator()
-                .filter(|comp| *comp != &SegmentComponent::TempStore)
-                .map(|component| self.relative_path(*component))
-                .collect::<HashSet<PathBuf>>()
-        }
+        SegmentComponent::iterator()
+            .map(|component| self.relative_path(*component))
+            .collect::<HashSet<PathBuf>>()
     }
 
     /// Returns the relative path of a component of our segment.
@@ -138,7 +115,6 @@ impl SegmentMeta {
             SegmentComponent::Positions => ".pos".to_string(),
             SegmentComponent::Terms => ".term".to_string(),
             SegmentComponent::Store => ".store".to_string(),
-            SegmentComponent::TempStore => ".store.temp".to_string(),
             SegmentComponent::FastFields => ".fast".to_string(),
             SegmentComponent::FieldNorms => ".fieldnorm".to_string(),
             SegmentComponent::Delete => format!(".{}.del", self.delete_opstamp().unwrap_or(0)),
@@ -183,7 +159,6 @@ impl SegmentMeta {
             segment_id: inner_meta.segment_id,
             max_doc,
             deletes: None,
-            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
         });
         SegmentMeta { tracked }
     }
@@ -202,7 +177,6 @@ impl SegmentMeta {
         let tracked = self.tracked.map(move |inner_meta| InnerSegmentMeta {
             segment_id: inner_meta.segment_id,
             max_doc: inner_meta.max_doc,
-            include_temp_doc_store: Arc::new(AtomicBool::new(true)),
             deletes: Some(delete_meta),
         });
         SegmentMeta { tracked }
@@ -214,14 +188,6 @@ struct InnerSegmentMeta {
     segment_id: SegmentId,
     max_doc: u32,
     pub deletes: Option<DeleteMeta>,
-    /// If you want to avoid the SegmentComponent::TempStore file to be covered by
-    /// garbage collection and deleted, set this to true. This is used during merge.
-    #[serde(skip)]
-    #[serde(default = "default_temp_store")]
-    pub(crate) include_temp_doc_store: Arc<AtomicBool>,
-}
-fn default_temp_store() -> Arc<AtomicBool> {
-    Arc::new(AtomicBool::new(false))
 }
 
 impl InnerSegmentMeta {

src/index/segment_component.rs

@@ -23,8 +23,6 @@ pub enum SegmentComponent {
     /// Accessing a document from the store is relatively slow, as it
     /// requires to decompress the entire block it belongs to.
     Store,
-    /// Temporary storage of the documents, before streamed to `Store`.
-    TempStore,
     /// Bitset describing which document of the segment is alive.
     /// (It was representing deleted docs but changed to represent alive docs from v0.17)
     Delete,
@@ -33,14 +31,13 @@ pub enum SegmentComponent {
 impl SegmentComponent {
     /// Iterates through the components.
     pub fn iterator() -> slice::Iter<'static, SegmentComponent> {
-        static SEGMENT_COMPONENTS: [SegmentComponent; 8] = [
+        static SEGMENT_COMPONENTS: [SegmentComponent; 7] = [
             SegmentComponent::Postings,
             SegmentComponent::Positions,
             SegmentComponent::FastFields,
             SegmentComponent::FieldNorms,
             SegmentComponent::Terms,
             SegmentComponent::Store,
-            SegmentComponent::TempStore,
             SegmentComponent::Delete,
         ];
         SEGMENT_COMPONENTS.iter()

src/indexer/index_writer.rs

@@ -218,7 +218,7 @@ fn index_documents<D: Document>(
     let alive_bitset_opt = apply_deletes(&segment_with_max_doc, &mut delete_cursor, &doc_opstamps)?;
 
     let meta = segment_with_max_doc.meta().clone();
-    meta.untrack_temp_docstore();
+
     // update segment_updater inventory to remove tempstore
     let segment_entry = SegmentEntry::new(meta, delete_cursor, alive_bitset_opt);
     segment_updater.schedule_add_segment(segment_entry).wait()?;

src/postings/block_segment_postings.rs

@@ -303,10 +303,10 @@ impl BlockSegmentPostings {
     }
 
     pub(crate) fn load_block(&mut self) {
-        let offset = self.skip_reader.byte_offset();
         if self.block_is_loaded() {
             return;
         }
+        let offset = self.skip_reader.byte_offset();
         match self.skip_reader.block_info() {
             BlockInfo::BitPacked {
                 doc_num_bits,

src/postings/segment_postings.rs

@@ -168,12 +168,20 @@ impl DocSet for SegmentPostings {
         self.doc()
     }
 
+    #[inline]
     fn seek(&mut self, target: DocId) -> DocId {
         debug_assert!(self.doc() <= target);
         if self.doc() >= target {
             return self.doc();
         }
 
+        // As an optimization, if the block is already loaded, we can
+        // cheaply check the next doc.
+        self.cur = (self.cur + 1).min(COMPRESSION_BLOCK_SIZE - 1);
+        if self.doc() >= target {
+            return self.doc();
+        }
+
         // Delegate block-local search to BlockSegmentPostings::seek, which returns
         // the in-block index of the first doc >= target.
         self.cur = self.block_cursor.seek(target);

src/query/boost_query.rs

@@ -1,6 +1,6 @@
 use std::fmt;
 
-use crate::docset::COLLECT_BLOCK_BUFFER_LEN;
+use crate::docset::{SeekDangerResult, COLLECT_BLOCK_BUFFER_LEN};
 use crate::fastfield::AliveBitSet;
 use crate::query::{EnableScoring, Explanation, Query, Scorer, Weight};
 use crate::{DocId, DocSet, Score, SegmentReader, Term};
@@ -104,8 +104,8 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
     fn seek(&mut self, target: DocId) -> DocId {
         self.underlying.seek(target)
     }
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        self.underlying.seek_into_the_danger_zone(target)
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        self.underlying.seek_danger(target)
     }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId; COLLECT_BLOCK_BUFFER_LEN]) -> usize {

src/query/disjunction.rs

@@ -1,6 +1,7 @@
 use std::cmp::Ordering;
 use std::collections::BinaryHeap;
 
+use crate::docset::SeekDangerResult;
 use crate::query::score_combiner::DoNothingCombiner;
 use crate::query::{ScoreCombiner, Scorer};
 use crate::{DocId, DocSet, Score, TERMINATED};
@@ -67,10 +68,12 @@ impl<T: Scorer> DocSet for ScorerWrapper<T> {
         self.current_doc = doc_id;
         doc_id
     }
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        let found = self.scorer.seek_into_the_danger_zone(target);
-        self.current_doc = self.scorer.doc();
-        found
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        let result = self.scorer.seek_danger(target);
+        if result == SeekDangerResult::Found {
+            self.current_doc = target;
+        }
+        result
     }
 
     fn doc(&self) -> DocId {

src/query/intersection.rs

@@ -1,5 +1,5 @@
 use super::size_hint::estimate_intersection;
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SeekDangerResult, TERMINATED};
 use crate::query::term_query::TermScorer;
 use crate::query::{EmptyScorer, Scorer};
 use crate::{DocId, Score};
@@ -84,6 +84,14 @@ impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
         docsets.sort_by_key(|docset| docset.cost());
         go_to_first_doc(&mut docsets);
         let left = docsets.remove(0);
+        debug_assert!({
+            let doc = left.doc();
+            if doc == TERMINATED {
+                true
+            } else {
+                docsets.iter().all(|docset| docset.doc() == doc)
+            }
+        });
         let right = docsets.remove(0);
         Intersection {
             left,
@@ -108,46 +116,61 @@ impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOt
     #[inline]
     fn advance(&mut self) -> DocId {
         let (left, right) = (&mut self.left, &mut self.right);
-        let mut candidate = left.advance();
-        if candidate == TERMINATED {
-            return TERMINATED;
-        }
 
-        loop {
-            // In the first part we look for a document in the intersection
-            // of the two rarest `DocSet` in the intersection.
+        // Invariant:
+        // - candidate is always <= to the next document in the intersection.
+        // - candidate strictly increases at every occurence of the loop.
+        let mut candidate = left.doc() + 1;
 
-            loop {
-                if right.seek_into_the_danger_zone(candidate) {
-                    break;
-                }
-                let right_doc = right.doc();
-                // TODO: Think about which value would make sense here
-                // It depends on the DocSet implementation, when a seek would outweigh an advance.
-                if right_doc > candidate.wrapping_add(100) {
-                    candidate = left.seek(right_doc);
-                } else {
-                    candidate = left.advance();
-                }
-                if candidate == TERMINATED {
-                    return TERMINATED;
-                }
-            }
+        // Termination: candidate strictly increases.
+        'outer: while candidate < TERMINATED {
+            // As we enter the loop, we should always have candidate < next_doc.
 
-            debug_assert_eq!(left.doc(), right.doc());
-            // test the remaining scorers
-            if self
-                .others
-                .iter_mut()
-                .all(|docset| docset.seek_into_the_danger_zone(candidate))
+            candidate = left.seek(candidate);
+
+            // Left is positionned on `candidate`.
+            debug_assert_eq!(left.doc(), candidate);
+
+            if let SeekDangerResult::SeekLowerBound(seek_lower_bound) = right.seek_danger(candidate)
             {
-                debug_assert_eq!(candidate, self.left.doc());
-                debug_assert_eq!(candidate, self.right.doc());
-                debug_assert!(self.others.iter().all(|docset| docset.doc() == candidate));
-                return candidate;
+                debug_assert!(
+                    seek_lower_bound == TERMINATED || seek_lower_bound > candidate,
+                    "seek_lower_bound {seek_lower_bound} must be greater than candidate \
+                     {candidate}"
+                );
+                candidate = seek_lower_bound;
+                continue;
             }
-            candidate = left.advance();
+
+            // Left and right are positionned on `candidate`.
+            debug_assert_eq!(right.doc(), candidate);
+
+            for other in &mut self.others {
+                if let SeekDangerResult::SeekLowerBound(seek_lower_bound) =
+                    other.seek_danger(candidate)
+                {
+                    // One of the scorer does not match, let's restart at the top of the loop.
+                    debug_assert!(
+                        seek_lower_bound == TERMINATED || seek_lower_bound > candidate,
+                        "seek_lower_bound {seek_lower_bound} must be greater than candidate \
+                         {candidate}"
+                    );
+                    candidate = seek_lower_bound;
+                    continue 'outer;
+                }
+            }
+
+            // At this point all scorers are in a valid state, aligned on the next document in the
+            // intersection.
+            debug_assert!(self.others.iter().all(|docset| docset.doc() == candidate));
+            return candidate;
         }
+
+        // We make sure our docset is in a valid state.
+        // In particular, we want .doc() to return TERMINATED.
+        left.seek(TERMINATED);
+
+        TERMINATED
     }
 
     fn seek(&mut self, target: DocId) -> DocId {
@@ -166,13 +189,19 @@ impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOt
     ///
     /// Some implementations may choose to advance past the target if beneficial for performance.
     /// The return value is `true` if the target is in the docset, and `false` otherwise.
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        self.left.seek_into_the_danger_zone(target)
-            && self.right.seek_into_the_danger_zone(target)
-            && self
-                .others
-                .iter_mut()
-                .all(|docset| docset.seek_into_the_danger_zone(target))
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        if let SeekDangerResult::SeekLowerBound(new_target) = self.left.seek_danger(target) {
+            return SeekDangerResult::SeekLowerBound(new_target);
+        }
+        if let SeekDangerResult::SeekLowerBound(new_target) = self.right.seek_danger(target) {
+            return SeekDangerResult::SeekLowerBound(new_target);
+        }
+        for docset in &mut self.others {
+            if let SeekDangerResult::SeekLowerBound(new_target) = docset.seek_danger(target) {
+                return SeekDangerResult::SeekLowerBound(new_target);
+            }
+        }
+        SeekDangerResult::Found
     }
 
     #[inline]
@@ -215,9 +244,12 @@ mod tests {
     use proptest::prelude::*;
 
     use super::Intersection;
+    use crate::collector::Count;
     use crate::docset::{DocSet, TERMINATED};
     use crate::postings::tests::test_skip_against_unoptimized;
-    use crate::query::VecDocSet;
+    use crate::query::{QueryParser, VecDocSet};
+    use crate::schema::{Schema, TEXT};
+    use crate::Index;
 
     #[test]
     fn test_intersection() {
@@ -304,6 +336,58 @@ mod tests {
         assert_eq!(intersection.doc(), TERMINATED);
     }
 
+    #[test]
+    fn test_intersection_abc() {
+        let a = VecDocSet::from(vec![2, 3, 6]);
+        let b = VecDocSet::from(vec![1, 3, 5]);
+        let c = VecDocSet::from(vec![1, 3, 5]);
+        let mut intersection = Intersection::new(vec![c, b, a], 10);
+        let mut docs = Vec::new();
+        use crate::DocSet;
+        while intersection.doc() != TERMINATED {
+            docs.push(intersection.doc());
+            intersection.advance();
+        }
+        assert_eq!(&docs, &[3]);
+    }
+
+    #[test]
+    fn test_intersection_termination() {
+        use crate::query::score_combiner::DoNothingCombiner;
+        use crate::query::{BufferedUnionScorer, ConstScorer, VecDocSet};
+
+        let a1 = ConstScorer::new(VecDocSet::from(vec![0u32, 10000]), 1.0);
+        let a2 = ConstScorer::new(VecDocSet::from(vec![0u32, 10000]), 1.0);
+
+        let mut b_scorers = vec![];
+        for _ in 0..2 {
+            // Union matches 0 and 10000.
+            b_scorers.push(ConstScorer::new(VecDocSet::from(vec![0, 10000]), 1.0));
+        }
+        // That's the union of two scores matching 0, and 10_000.
+        let union = BufferedUnionScorer::build(b_scorers, DoNothingCombiner::default, 30000);
+
+        // Mismatching scorer: matches 0 and 20000. We then append more docs at the end to ensure it
+        // is last.
+        let mut m_docs = vec![0, 20000];
+        for i in 30000..30100 {
+            m_docs.push(i);
+        }
+        let m = ConstScorer::new(VecDocSet::from(m_docs), 1.0);
+
+        // Costs: A1=2, A2=2, Union=4, M=102.
+        // Sorted: A1, A2, Union, M.
+        // Left=A1, Right=A2, Others=[Union, M].
+        let mut intersection = crate::query::intersect_scorers(
+            vec![Box::new(a1), Box::new(a2), Box::new(union), Box::new(m)],
+            40000,
+        );
+
+        while intersection.doc() != TERMINATED {
+            intersection.advance();
+        }
+    }
+
     // Strategy to generate sorted and deduplicated vectors of u32 document IDs
     fn sorted_deduped_vec(max_val: u32, max_size: usize) -> impl Strategy<Value = Vec<u32>> {
         prop::collection::vec(0..max_val, 0..max_size).prop_map(|mut vec| {
@@ -335,6 +419,30 @@ mod tests {
             }
             assert_eq!(intersection.doc(), TERMINATED);
         }
+    }
 
+    #[test]
+    fn test_bug_2811_intersection_candidate_should_increase() {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let schema = schema_builder.build();
+
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+        writer
+            .add_document(doc!(text_field=>"hello happy tax"))
+            .unwrap();
+        writer.add_document(doc!(text_field=>"hello")).unwrap();
+        writer.add_document(doc!(text_field=>"hello")).unwrap();
+        writer.add_document(doc!(text_field=>"happy tax")).unwrap();
+
+        writer.commit().unwrap();
+        let query_parser = QueryParser::for_index(&index, Vec::new());
+        let query = query_parser
+            .parse_query(r#"+text:hello +text:"happy tax""#)
+            .unwrap();
+        let searcher = index.reader().unwrap().searcher();
+        let c = searcher.search(&*query, &Count).unwrap();
+        assert_eq!(c, 1);
     }
 }

src/query/phrase_prefix_query/phrase_prefix_scorer.rs

@@ -1,4 +1,4 @@
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SeekDangerResult, TERMINATED};
 use crate::fieldnorm::FieldNormReader;
 use crate::postings::Postings;
 use crate::query::bm25::Bm25Weight;
@@ -194,11 +194,16 @@ impl<TPostings: Postings> DocSet for PhrasePrefixScorer<TPostings> {
         self.advance()
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        if self.phrase_scorer.seek_into_the_danger_zone(target) {
-            self.matches_prefix()
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        let seek_res = self.phrase_scorer.seek_danger(target);
+        if seek_res != SeekDangerResult::Found {
+            return seek_res;
+        }
+        // The intersection matched. Now let's see if we match the prefix.
+        if self.matches_prefix() {
+            SeekDangerResult::Found
         } else {
-            false
+            SeekDangerResult::SeekLowerBound(target + 1)
         }
     }
 

src/query/phrase_query/phrase_scorer.rs

@@ -1,6 +1,6 @@
 use std::cmp::Ordering;
 
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SeekDangerResult, TERMINATED};
 use crate::fieldnorm::FieldNormReader;
 use crate::postings::Postings;
 use crate::query::bm25::Bm25Weight;
@@ -530,12 +530,23 @@ impl<TPostings: Postings> DocSet for PhraseScorer<TPostings> {
         self.advance()
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
-        debug_assert!(target >= self.doc());
-        if self.intersection_docset.seek_into_the_danger_zone(target) && self.phrase_match() {
-            return true;
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        debug_assert!(
+            target >= self.doc(),
+            "target ({}) should be greater than or equal to doc ({})",
+            target,
+            self.doc()
+        );
+        let seek_res = self.intersection_docset.seek_danger(target);
+        if seek_res != SeekDangerResult::Found {
+            return seek_res;
+        }
+        // The intersection matched. Now let's see if we match the phrase.
+        if self.phrase_match() {
+            SeekDangerResult::Found
+        } else {
+            SeekDangerResult::SeekLowerBound(target + 1)
         }
-        false
     }
 
     fn doc(&self) -> DocId {

src/query/reqopt_scorer.rs

@@ -1,6 +1,6 @@
 use std::marker::PhantomData;
 
-use crate::docset::DocSet;
+use crate::docset::{DocSet, SeekDangerResult};
 use crate::query::score_combiner::ScoreCombiner;
 use crate::query::Scorer;
 use crate::{DocId, Score};
@@ -56,9 +56,9 @@ where
         self.req_scorer.seek(target)
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
         self.score_cache = None;
-        self.req_scorer.seek_into_the_danger_zone(target)
+        self.req_scorer.seek_danger(target)
     }
 
     fn doc(&self) -> DocId {

src/query/term_query/term_scorer.rs

@@ -105,6 +105,7 @@ impl DocSet for TermScorer {
 
     #[inline]
     fn seek(&mut self, target: DocId) -> DocId {
+        debug_assert!(target >= self.doc());
         self.postings.seek(target)
     }
 

src/query/union/buffered_union.rs

@@ -1,6 +1,6 @@
 use common::TinySet;
 
-use crate::docset::{DocSet, TERMINATED};
+use crate::docset::{DocSet, SeekDangerResult, TERMINATED};
 use crate::query::score_combiner::{DoNothingCombiner, ScoreCombiner};
 use crate::query::size_hint::estimate_union;
 use crate::query::Scorer;
@@ -225,25 +225,47 @@ where
         }
     }
 
-    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+    fn seek_danger(&mut self, target: DocId) -> SeekDangerResult {
+        if target >= TERMINATED {
+            return SeekDangerResult::SeekLowerBound(TERMINATED);
+        }
         if self.is_in_horizon(target) {
             // Our value is within the buffered horizon and the docset may already have been
             // processed and removed, so we need to use seek, which uses the regular advance.
-            self.seek(target) == target
-        } else {
-            // The docsets are not in the buffered range, so we can use seek_into_the_danger_zone
-            // of the underlying docsets
-            let is_hit = self
-                .docsets
-                .iter_mut()
-                .any(|docset| docset.seek_into_the_danger_zone(target));
+            let seek_doc = self.seek(target);
+            if seek_doc == target {
+                return SeekDangerResult::Found;
+            } else {
+                return SeekDangerResult::SeekLowerBound(seek_doc);
+            };
+        }
 
-            // The API requires the DocSet to be in a valid state when `seek_into_the_danger_zone`
-            // returns true.
-            if is_hit {
-                self.seek(target);
+        // The docsets are not in the buffered range, so we can use seek_into_the_danger_zone
+        // of the underlying docsets
+        let mut is_hit = false;
+        let mut min_new_target = TERMINATED;
+
+        for docset in self.docsets.iter_mut() {
+            match docset.seek_danger(target) {
+                SeekDangerResult::Found => {
+                    is_hit = true;
+                    break;
+                }
+                SeekDangerResult::SeekLowerBound(new_target) => {
+                    min_new_target = min_new_target.min(new_target);
+                }
             }
-            is_hit
+        }
+
+        // The API requires the DocSet to be in a valid state when `seek_into_the_danger_zone`
+        // returns Found.
+        if is_hit {
+            // The doc is found. Let's make sure we position the union on the target
+            // to bring it back to a valid state.
+            self.seek(target);
+            SeekDangerResult::Found
+        } else {
+            SeekDangerResult::SeekLowerBound(min_new_target)
         }
     }
 

src/query/union/mod.rs

@@ -14,7 +14,7 @@ mod tests {
     use common::BitSet;
 
     use super::{SimpleUnion, *};
-    use crate::docset::{DocSet, TERMINATED};
+    use crate::docset::{DocSet, SeekDangerResult, TERMINATED};
     use crate::postings::tests::test_skip_against_unoptimized;
     use crate::query::score_combiner::DoNothingCombiner;
     use crate::query::union::bitset_union::BitSetPostingUnion;
@@ -254,6 +254,27 @@ mod tests {
             vec![1, 2, 3, 7, 8, 9, 99, 100, 101, 500, 20000],
         );
     }
+
+    #[test]
+    fn test_buffered_union_seek_into_danger_zone_terminated() {
+        let scorer1 = ConstScorer::new(VecDocSet::from(vec![1, 2]), 1.0);
+        let scorer2 = ConstScorer::new(VecDocSet::from(vec![2, 3]), 1.0);
+
+        let mut union_scorer =
+            BufferedUnionScorer::build(vec![scorer1, scorer2], DoNothingCombiner::default, 100);
+
+        // Advance to end
+        while union_scorer.doc() != TERMINATED {
+            union_scorer.advance();
+        }
+
+        assert_eq!(union_scorer.doc(), TERMINATED);
+
+        assert_eq!(
+            union_scorer.seek_danger(TERMINATED),
+            SeekDangerResult::SeekLowerBound(TERMINATED)
+        );
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]

src/query/vec_docset.rs

@@ -17,6 +17,9 @@ pub struct VecDocSet {
 
 impl From<Vec<DocId>> for VecDocSet {
     fn from(doc_ids: Vec<DocId>) -> VecDocSet {
+        // We do not use `slice::is_sorted`, as we want to check for doc ids to be strictly
+        // sorted.
+        assert!(doc_ids.windows(2).all(|w| w[0] < w[1]));
         VecDocSet { doc_ids, cursor: 0 }
     }
 }

src/space_usage/mod.rs

@@ -124,7 +124,6 @@ impl SegmentSpaceUsage {
             FieldNorms => PerField(self.fieldnorms().clone()),
             Terms => PerField(self.termdict().clone()),
             SegmentComponent::Store => ComponentSpaceUsage::Store(self.store().clone()),
-            SegmentComponent::TempStore => ComponentSpaceUsage::Store(self.store().clone()),
             Delete => Basic(self.deletes()),
         }
     }