Truncate facet values that are too long before indexing them

milli/src/update/index_documents/extract/extract_facet_string_docids.rs

@@ -6,7 +6,8 @@ use heed::BytesEncode;
 use super::helpers::{create_sorter, sorter_into_reader, try_split_array_at, GrenadParameters};
 use crate::heed_codec::facet::{FacetGroupKey, FacetGroupKeyCodec};
 use crate::heed_codec::StrRefCodec;
-use crate::update::index_documents::{merge_cbo_roaring_bitmaps, valid_lmdb_key};
+use crate::update::index_documents::helpers::MAX_FACET_VALUE_LENGTH;
+use crate::update::index_documents::merge_cbo_roaring_bitmaps;
 use crate::{FieldId, Result};
 
 /// Extracts the facet string and the documents ids where this facet string appear.
@@ -38,14 +39,22 @@ pub fn extract_facet_string_docids<R: io::Read + io::Seek>(
             try_split_array_at::<_, 4>(bytes).unwrap();
         let document_id = u32::from_be_bytes(document_id_bytes);
 
-        let normalised_value = std::str::from_utf8(normalized_value_bytes)?;
+        let mut normalised_value = std::str::from_utf8(normalized_value_bytes)?;
+
+        let normalised_truncated_value: String;
+        if normalised_value.len() > MAX_FACET_VALUE_LENGTH {
+            normalised_truncated_value = normalised_value
+                .char_indices()
+                .take_while(|(idx, _)| idx + 4 < MAX_FACET_VALUE_LENGTH)
+                .map(|(_, c)| c)
+                .collect();
+            normalised_value = normalised_truncated_value.as_str();
+        }
         let key = FacetGroupKey { field_id, level: 0, left_bound: normalised_value };
         let key_bytes = FacetGroupKeyCodec::<StrRefCodec>::bytes_encode(&key).unwrap();
-        if valid_lmdb_key(&key_bytes) {
         // document id is encoded in native-endian because of the CBO roaring bitmap codec
         facet_string_docids_sorter.insert(&key_bytes, document_id.to_ne_bytes())?;
     }
-    }
 
     sorter_into_reader(facet_string_docids_sorter, indexer)
 }

milli/src/update/index_documents/extract/extract_fid_docid_facet_values.rs

@@ -12,6 +12,7 @@ use serde_json::Value;
 use super::helpers::{create_sorter, keep_first, sorter_into_reader, GrenadParameters};
 use crate::error::InternalError;
 use crate::facet::value_encoding::f64_into_bytes;
+use crate::update::index_documents::helpers::MAX_FACET_VALUE_LENGTH;
 use crate::update::index_documents::{create_writer, writer_into_reader};
 use crate::{CboRoaringBitmapCodec, DocumentId, FieldId, Result, BEU32};
 
@@ -87,8 +88,14 @@ pub fn extract_fid_docid_facet_values<R: io::Read + io::Seek>(
 
                 // insert normalized and original facet string in sorter
                 for (normalized, original) in strings.into_iter().filter(|(n, _)| !n.is_empty()) {
+                    let normalised_truncated_value: String = normalized
+                        .char_indices()
+                        .take_while(|(idx, _)| idx + 4 < MAX_FACET_VALUE_LENGTH)
+                        .map(|(_, c)| c)
+                        .collect();
+
                     key_buffer.truncate(size_of::<FieldId>() + size_of::<DocumentId>());
-                    key_buffer.extend_from_slice(normalized.as_bytes());
+                    key_buffer.extend_from_slice(normalised_truncated_value.as_bytes());
                     fid_docid_facet_strings_sorter.insert(&key_buffer, original.as_bytes())?;
                 }
             }

milli/src/update/index_documents/helpers/mod.rs

@@ -18,8 +18,20 @@ pub use merge_functions::{
     serialize_roaring_bitmap, MergeFn,
 };
 
+/// The maximum length a LMDB key can be.
+///
+/// Note that the actual allowed length is a little bit higher, but
+/// we keep a margin of safety.
+const MAX_LMDB_KEY_LENGTH: usize = 500;
+
+/// The maximum length a field value can be when inserted in an LMDB key.
+///
+/// This number is determined by the keys of the different facet databases
+/// and adding a margin of safety.
+pub const MAX_FACET_VALUE_LENGTH: usize = MAX_LMDB_KEY_LENGTH - 20;
+
 /// The maximum length a word can be
-pub const MAX_WORD_LENGTH: usize = 250;
+pub const MAX_WORD_LENGTH: usize = MAX_LMDB_KEY_LENGTH / 2;
 
 pub fn valid_lmdb_key(key: impl AsRef<[u8]>) -> bool {
     key.as_ref().len() <= MAX_WORD_LENGTH * 2 && !key.as_ref().is_empty()