diff --git a/Cargo.toml b/Cargo.toml
index dadb7ee99..21b412b87 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -108,7 +108,7 @@ unstable = [] # useful for benches.
 quickwit = ["sstable"]
 
 [workspace]
-members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes", "stacker", "sstable"]
+members = ["query-grammar", "bitpacker", "common", "fastfield_codecs", "ownedbytes", "stacker", "sstable", "columnar"]
 
 # Following the "fail" crate best practises, we isolate
 # tests that define specific behavior in fail check points
diff --git a/columnar/Cargo.toml b/columnar/Cargo.toml
new file mode 100644
index 000000000..55ff5c761
--- /dev/null
+++ b/columnar/Cargo.toml
@@ -0,0 +1,26 @@
+[package]
+name = "tantivy-columnar"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]
+stacker = { path = "../stacker", package="tantivy-stacker"}
+serde_json = "1"
+thiserror = "1"
+fnv = "1"
+tantivy-fst =  "0.4.0"
+sstable = { path = "../sstable", package = "tantivy-sstable" }
+common = { path = "../common", package = "tantivy-common" }
+fastfield_codecs = { path = "../fastfield_codecs"}
+ordered-float = "3.4"
+itertools = "0.10"
+
+[features]
+# default = ["quickwit"]
+# quickwit = ["common/quickwit"]
+
+
+
+
+[dev-dependencies]
+proptest = "1"
diff --git a/columnar/README.md b/columnar/README.md
new file mode 100644
index 000000000..35a4f778d
--- /dev/null
+++ b/columnar/README.md
@@ -0,0 +1,33 @@
+# Columnar format
+
+This crate describes columnar format used in tantivy.
+
+
+## Goals
+
+This format is special in the following way.
+- it needs to be compact
+- it does not required to be loaded in memory.
+- it is designed to fit well with quickwit's strange constraint:
+we need to be able to load columns rapidly.
+- columns of several types can be associated with the same column name.
+- it needs to support columns with different types `(str, u64, i64, f64)`
+and different cardinality `(required, optional, multivalued)`.
+- columns, once loaded, offer cheap random access.
+
+# Format
+
+A quickwit/tantivy style sstable associated
+`(column names, column_cardinality, column_type) to range of bytes.
+
+The format of the key is:
+`[column_name][ZERO_BYTE][column_type_header: u8]`
+
+Column name may not contain the zero byte.
+
+Listing all columns associated to `column_name` can therefore
+be done by listing all keys prefixed by
+`[column_name][ZERO_BYTE]`
+
+The associated range of bytes refer to a range of bytes
+
diff --git a/columnar/src/column_type_header.rs b/columnar/src/column_type_header.rs
new file mode 100644
index 000000000..769557eff
--- /dev/null
+++ b/columnar/src/column_type_header.rs
@@ -0,0 +1,154 @@
+use crate::value::NumericalType;
+
+#[derive(Clone, Copy, Hash, Default, Debug, PartialEq, Eq, PartialOrd, Ord)]
+#[repr(u8)]
+pub enum Cardinality {
+    #[default]
+    Required = 0,
+    Optional = 1,
+    Multivalued = 2,
+}
+
+impl Cardinality {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn try_from_code(code: u8) -> Option<Cardinality> {
+        match code {
+            0 => Some(Cardinality::Required),
+            1 => Some(Cardinality::Optional),
+            2 => Some(Cardinality::Multivalued),
+            _ => None,
+        }
+    }
+}
+
+#[derive(Hash, Eq, PartialEq, Debug, Clone, Copy)]
+pub enum ColumnType {
+    Bytes,
+    Numerical(NumericalType),
+}
+
+impl ColumnType {
+    pub fn to_code(self) -> u8 {
+        match self {
+            ColumnType::Bytes => 0u8,
+            ColumnType::Numerical(numerical_type) => 1u8 | (numerical_type.to_code() << 1),
+        }
+    }
+
+    pub fn try_from_code(code: u8) -> Option<ColumnType> {
+        if code == 0u8 {
+            return Some(ColumnType::Bytes);
+        }
+        if code & 1u8 == 0u8 {
+            return None;
+        }
+        let numerical_type = NumericalType::try_from_code(code >> 1)?;
+        Some(ColumnType::Numerical(numerical_type))
+    }
+}
+
+/// Represents the type and cardinality of a column.
+/// This is encoded over one-byte and added to a column key in the
+/// columnar sstable.
+///
+/// Cardinality is encoded as the first two highest two bits.
+/// The low 6 bits encode the column type.
+#[derive(Eq, Hash, PartialEq, Debug, Copy, Clone)]
+pub struct ColumnTypeAndCardinality {
+    pub cardinality: Cardinality,
+    pub typ: ColumnType,
+}
+
+#[inline]
+const fn compute_mask(num_bits: u8) -> u8 {
+    if num_bits == 8 {
+        u8::MAX
+    } else {
+        (1u8 << num_bits) - 1
+    }
+}
+
+#[inline]
+fn select_bits<const START: u8, const END: u8>(code: u8) -> u8 {
+    assert!(START <= END);
+    assert!(END <= 8);
+    let num_bits: u8 = END - START;
+    let mask: u8 = compute_mask(num_bits);
+    (code >> START) & mask
+}
+
+#[inline]
+fn place_bits<const START: u8, const END: u8>(code: u8) -> u8 {
+    assert!(START <= END);
+    assert!(END <= 8);
+    let num_bits: u8 = END - START;
+    let mask: u8 = compute_mask(num_bits);
+    assert!(code <= mask);
+    code << START
+}
+
+impl ColumnTypeAndCardinality {
+    pub fn to_code(self) -> u8 {
+        place_bits::<6, 8>(self.cardinality.to_code()) | place_bits::<0, 6>(self.typ.to_code())
+    }
+
+    pub fn try_from_code(code: u8) -> Option<ColumnTypeAndCardinality> {
+        let typ_code = select_bits::<0, 6>(code);
+        let cardinality_code = select_bits::<6, 8>(code);
+        let cardinality = Cardinality::try_from_code(cardinality_code)?;
+        let typ = ColumnType::try_from_code(typ_code)?;
+        assert_eq!(typ.to_code(), typ_code);
+        Some(ColumnTypeAndCardinality { cardinality, typ })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::HashSet;
+
+    use super::ColumnTypeAndCardinality;
+    use crate::column_type_header::{Cardinality, ColumnType};
+
+    #[test]
+    fn test_column_type_header_to_code() {
+        let mut column_type_header_set: HashSet<ColumnTypeAndCardinality> = HashSet::new();
+        for code in u8::MIN..=u8::MAX {
+            if let Some(column_type_header) = ColumnTypeAndCardinality::try_from_code(code) {
+                assert_eq!(column_type_header.to_code(), code);
+                assert!(column_type_header_set.insert(column_type_header));
+            }
+        }
+        assert_eq!(
+            column_type_header_set.len(),
+            3 /* cardinality */ * (1 + 3) // column_types
+        );
+    }
+
+    #[test]
+    fn test_column_type_to_code() {
+        let mut column_type_set: HashSet<ColumnType> = HashSet::new();
+        for code in u8::MIN..=u8::MAX {
+            if let Some(column_type) = ColumnType::try_from_code(code) {
+                assert_eq!(column_type.to_code(), code);
+                assert!(column_type_set.insert(column_type));
+            }
+        }
+        assert_eq!(column_type_set.len(), 1 + 3);
+    }
+
+    #[test]
+    fn test_cardinality_to_code() {
+        let mut num_cardinality = 0;
+        for code in u8::MIN..=u8::MAX {
+            let cardinality_opt = Cardinality::try_from_code(code);
+            if let Some(cardinality) = cardinality_opt {
+                assert_eq!(cardinality.to_code(), code);
+                num_cardinality += 1;
+            }
+        }
+        assert_eq!(num_cardinality, 3);
+    }
+}
diff --git a/columnar/src/dictionary.rs b/columnar/src/dictionary.rs
new file mode 100644
index 000000000..11754b7b6
--- /dev/null
+++ b/columnar/src/dictionary.rs
@@ -0,0 +1,78 @@
+use std::io;
+
+use fnv::FnvHashMap;
+
+fn fst_err_into_io_err(fst_err: tantivy_fst::Error) -> io::Error {
+    match fst_err {
+        tantivy_fst::Error::Fst(fst_err) => {
+            io::Error::new(io::ErrorKind::Other, format!("FST Error: {:?}", fst_err))
+        }
+        tantivy_fst::Error::Io(io_err) => io_err,
+    }
+}
+
+/// `DictionaryBuilder` for dictionary encoding.
+///
+/// It stores the different terms encounterred and assigns them a temporary value
+/// we call unordered id.
+///
+/// Upon serialization, we will sort the ids and hence build a `UnorderedId -> Term ordinal`
+/// mapping.
+#[derive(Default)]
+pub struct DictionaryBuilder {
+    dict: FnvHashMap<Vec<u8>, UnorderedId>,
+}
+
+pub struct IdMapping {
+    unordered_to_ord: Vec<OrderedId>,
+}
+
+impl IdMapping {
+    pub fn to_ord(&self, unordered: UnorderedId) -> OrderedId {
+        self.unordered_to_ord[unordered.0 as usize]
+    }
+}
+
+impl DictionaryBuilder {
+    /// Get or allocate an unordered id.
+    /// (This ID is simply an auto-incremented id.)
+    pub fn get_or_allocate_id(&mut self, term: &[u8]) -> UnorderedId {
+        if let Some(term_id) = self.dict.get(term) {
+            return *term_id;
+        }
+        let new_id = UnorderedId(self.dict.len() as u32);
+        self.dict.insert(term.to_vec(), new_id);
+        new_id
+    }
+
+    /// Serialize the dictionary into an fst, and returns the
+    /// `UnorderedId -> TermOrdinal` map.
+    pub fn serialize<'a, W: io::Write + 'a>(&self, wrt: &mut W) -> io::Result<IdMapping> {
+        serialize_inner(&self.dict, wrt).map_err(fst_err_into_io_err)
+    }
+}
+
+/// Helper function just there for error conversion.
+fn serialize_inner<'a, W: io::Write + 'a>(
+    dict: &FnvHashMap<Vec<u8>, UnorderedId>,
+    wrt: &mut W,
+) -> tantivy_fst::Result<IdMapping> {
+    let mut terms: Vec<(&[u8], UnorderedId)> =
+        dict.iter().map(|(k, v)| (k.as_slice(), *v)).collect();
+    terms.sort_unstable_by_key(|(key, _)| *key);
+    let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
+    let mut fst_builder = tantivy_fst::MapBuilder::new(wrt)?;
+    for (ord, (key, unordered_id)) in terms.into_iter().enumerate() {
+        let ordered_id = OrderedId(ord as u32);
+        fst_builder.insert(key, ord as u64)?;
+        unordered_to_ord[unordered_id.0 as usize] = ordered_id;
+    }
+    fst_builder.finish()?;
+    Ok(IdMapping { unordered_to_ord })
+}
+
+#[derive(Clone, Copy, Debug)]
+pub struct UnorderedId(pub u32);
+
+#[derive(Clone, Copy)]
+pub struct OrderedId(pub u32);
diff --git a/columnar/src/lib.rs b/columnar/src/lib.rs
new file mode 100644
index 000000000..89b5d2504
--- /dev/null
+++ b/columnar/src/lib.rs
@@ -0,0 +1,69 @@
+// Copyright (C) 2022 Quickwit, Inc.
+//
+// Quickwit is offered under the AGPL v3.0 and as commercial software.
+// For commercial licensing, contact us at hello@quickwit.io.
+//
+// AGPL:
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as
+// published by the Free Software Foundation, either version 3 of the
+// License, or (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+mod column_type_header;
+mod dictionary;
+mod reader;
+mod serializer;
+mod value;
+mod writer;
+
+pub use column_type_header::Cardinality;
+pub use reader::ColumnarReader;
+pub use serializer::ColumnarSerializer;
+pub use writer::ColumnarWriter;
+
+pub type DocId = u32;
+
+#[cfg(test)]
+mod tests {
+    use std::ops::Range;
+
+    use common::file_slice::FileSlice;
+
+    use crate::column_type_header::ColumnTypeAndCardinality;
+    use crate::reader::ColumnarReader;
+    use crate::serializer::ColumnarSerializer;
+    use crate::value::NumericalValue;
+    use crate::ColumnarWriter;
+
+    #[test]
+    fn test_dataframe_writer() {
+        let mut dataframe_writer = ColumnarWriter::default();
+        dataframe_writer.record_numerical(1u32, b"srical.value", NumericalValue::U64(1u64));
+        dataframe_writer.record_numerical(2u32, b"srical.value", NumericalValue::U64(2u64));
+        dataframe_writer.record_numerical(4u32, b"srical.value", NumericalValue::I64(2i64));
+        let mut buffer: Vec<u8> = Vec::new();
+        let serializer = ColumnarSerializer::new(&mut buffer);
+        dataframe_writer.serialize(5, serializer).unwrap();
+        let columnar_fileslice = FileSlice::from(buffer);
+        let columnar = ColumnarReader::open(columnar_fileslice).unwrap();
+        assert_eq!(columnar.num_columns(), 1);
+        let cols: Vec<(ColumnTypeAndCardinality, Range<u64>)> =
+            columnar.read_columns("srical.value").unwrap();
+        assert_eq!(cols.len(), 1);
+        // Right now this 31 bytes are spent as follows
+        //
+        // - header 14 bytes
+        // - vals  8 //< due to padding? could have been 1byte?.
+        // - null footer 6 bytes
+        // - version footer 3 bytes // Should be file-wide
+        assert_eq!(cols[0].1, 0..31);
+    }
+}
diff --git a/columnar/src/reader/mod.rs b/columnar/src/reader/mod.rs
new file mode 100644
index 000000000..b5dc1a1f9
--- /dev/null
+++ b/columnar/src/reader/mod.rs
@@ -0,0 +1,66 @@
+use std::ops::Range;
+use std::{io, mem};
+
+use common::file_slice::FileSlice;
+use common::BinarySerializable;
+use sstable::{Dictionary, SSTableRange};
+
+use crate::column_type_header::ColumnTypeAndCardinality;
+
+fn io_invalid_data(msg: String) -> io::Error {
+    io::Error::new(io::ErrorKind::InvalidData, msg) // format!("Invalid key found.
+                                                    // {key_bytes:?}")));
+}
+pub struct ColumnarReader {
+    column_dictionary: Dictionary<SSTableRange>,
+    column_data: FileSlice,
+}
+
+impl ColumnarReader {
+    pub fn num_columns(&self) -> usize {
+        self.column_dictionary.num_terms()
+    }
+
+    pub fn open(file_slice: FileSlice) -> io::Result<ColumnarReader> {
+        let (file_slice_without_sstable_len, sstable_len_bytes) =
+            file_slice.split_from_end(mem::size_of::<u64>());
+        let mut sstable_len_bytes = sstable_len_bytes.read_bytes()?;
+        let sstable_len = u64::deserialize(&mut sstable_len_bytes)?;
+        let (column_data, sstable) =
+            file_slice_without_sstable_len.split_from_end(sstable_len as usize);
+        let column_dictionary = Dictionary::open(sstable)?;
+        Ok(ColumnarReader {
+            column_dictionary,
+            column_data,
+        })
+    }
+
+    pub fn read_columns(
+        &self,
+        field_name: &str,
+    ) -> io::Result<Vec<(ColumnTypeAndCardinality, Range<u64>)>> {
+        let mut start_key = field_name.to_string();
+        start_key.push('\0');
+        let mut end_key = field_name.to_string();
+        end_key.push(1u8 as char);
+        let mut stream = self
+            .column_dictionary
+            .range()
+            .ge(start_key.as_bytes())
+            .lt(end_key.as_bytes())
+            .into_stream()?;
+        let mut results = Vec::new();
+        while stream.advance() {
+            let key_bytes: &[u8] = stream.key();
+            if !key_bytes.starts_with(start_key.as_bytes()) {
+                return Err(io_invalid_data(format!("Invalid key found. {key_bytes:?}")));
+            }
+            let column_code: u8 = key_bytes.last().cloned().unwrap();
+            let column_type_and_cardinality = ColumnTypeAndCardinality::try_from_code(column_code)
+                .ok_or_else(|| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
+            let range = stream.value().clone();
+            results.push((column_type_and_cardinality, range));
+        }
+        Ok(results)
+    }
+}
diff --git a/columnar/src/serializer.rs b/columnar/src/serializer.rs
new file mode 100644
index 000000000..a66b06585
--- /dev/null
+++ b/columnar/src/serializer.rs
@@ -0,0 +1,39 @@
+use std::io;
+use std::io::Write;
+use std::ops::Range;
+
+use common::CountingWriter;
+use sstable::value::RangeWriter;
+use sstable::SSTableRange;
+
+pub struct ColumnarSerializer<W: io::Write> {
+    wrt: CountingWriter<W>,
+    sstable_range: sstable::Writer<Vec<u8>, RangeWriter>,
+}
+
+impl<W: io::Write> ColumnarSerializer<W> {
+    pub fn new(wrt: W) -> ColumnarSerializer<W> {
+        let sstable_range: sstable::Writer<Vec<u8>, RangeWriter> =
+            sstable::Dictionary::<SSTableRange>::builder(Vec::with_capacity(100_000)).unwrap();
+        ColumnarSerializer {
+            wrt: CountingWriter::wrap(wrt),
+            sstable_range,
+        }
+    }
+
+    pub fn record_column_offsets(&mut self, key: &[u8], byte_range: Range<u64>) -> io::Result<()> {
+        self.sstable_range.insert(key, &byte_range)
+    }
+
+    pub fn wrt(&mut self) -> &mut CountingWriter<W> {
+        &mut self.wrt
+    }
+
+    pub fn finalize(mut self) -> io::Result<()> {
+        let sstable_bytes: Vec<u8> = self.sstable_range.finish()?;
+        let sstable_num_bytes: u64 = sstable_bytes.len() as u64;
+        self.wrt.write_all(&sstable_bytes)?;
+        self.wrt.write_all(&sstable_num_bytes.to_le_bytes()[..])?;
+        Ok(())
+    }
+}
diff --git a/columnar/src/value.rs b/columnar/src/value.rs
new file mode 100644
index 000000000..aa200a7e5
--- /dev/null
+++ b/columnar/src/value.rs
@@ -0,0 +1,123 @@
+use ordered_float::NotNan;
+
+#[derive(Copy, Clone, Debug, PartialEq)]
+pub enum NumericalValue {
+    I64(i64),
+    U64(u64),
+    F64(NotNan<f64>),
+}
+
+impl From<u64> for NumericalValue {
+    fn from(val: u64) -> NumericalValue {
+        NumericalValue::U64(val)
+    }
+}
+
+impl From<i64> for NumericalValue {
+    fn from(val: i64) -> Self {
+        NumericalValue::I64(val)
+    }
+}
+
+impl From<NotNan<f64>> for NumericalValue {
+    fn from(val: NotNan<f64>) -> Self {
+        NumericalValue::F64(val)
+    }
+}
+
+impl NumericalValue {
+    pub fn numerical_type(&self) -> NumericalType {
+        match self {
+            NumericalValue::F64(_) => NumericalType::F64,
+            NumericalValue::I64(_) => NumericalType::I64,
+            NumericalValue::U64(_) => NumericalType::U64,
+        }
+    }
+}
+
+impl Eq for NumericalValue {}
+
+#[derive(Clone, Copy, Debug, Default, Hash, Eq, PartialEq)]
+#[repr(u8)]
+pub enum NumericalType {
+    #[default]
+    I64 = 0,
+    U64 = 1,
+    F64 = 2,
+}
+
+impl NumericalType {
+    pub fn to_code(self) -> u8 {
+        self as u8
+    }
+
+    pub fn try_from_code(code: u8) -> Option<NumericalType> {
+        match code {
+            0 => Some(NumericalType::I64),
+            1 => Some(NumericalType::U64),
+            2 => Some(NumericalType::F64),
+            _ => None,
+        }
+    }
+}
+
+/// We voluntarily avoid using `Into` here to keep this
+/// implementation quirk as private as possible.
+///
+/// This coercion trait actually panics if it is used
+/// to convert a loose types to a stricter type.
+///
+/// The level is strictness is somewhat arbitrary.
+/// - i64
+/// - u64
+/// - f64.
+pub(crate) trait Coerce {
+    fn coerce(numerical_value: NumericalValue) -> Self;
+}
+
+impl Coerce for i64 {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => val,
+            NumericalValue::U64(val) => val as i64,
+            NumericalValue::F64(_) => unreachable!(),
+        }
+    }
+}
+
+impl Coerce for u64 {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => val as u64,
+            NumericalValue::U64(val) => val,
+            NumericalValue::F64(_) => unreachable!(),
+        }
+    }
+}
+
+impl Coerce for NotNan<f64> {
+    fn coerce(value: NumericalValue) -> Self {
+        match value {
+            NumericalValue::I64(val) => unsafe { NotNan::new_unchecked(val as f64) },
+            NumericalValue::U64(val) => unsafe { NotNan::new_unchecked(val as f64) },
+            NumericalValue::F64(val) => val,
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::NumericalType;
+
+    #[test]
+    fn test_numerical_type_code() {
+        let mut num_numerical_type = 0;
+        for code in u8::MIN..=u8::MAX {
+            if let Some(numerical_type) = NumericalType::try_from_code(code) {
+                assert_eq!(numerical_type.to_code(), code);
+                num_numerical_type += 1;
+            }
+        }
+        assert_eq!(num_numerical_type, 3);
+    }
+}
diff --git a/columnar/src/writer/column_operation.rs b/columnar/src/writer/column_operation.rs
new file mode 100644
index 000000000..e01842562
--- /dev/null
+++ b/columnar/src/writer/column_operation.rs
@@ -0,0 +1,321 @@
+use std::fmt;
+use std::num::NonZeroU8;
+
+use ordered_float::NotNan;
+use thiserror::Error;
+
+use crate::dictionary::UnorderedId;
+use crate::value::NumericalValue;
+use crate::DocId;
+
+/// When we build a columnar dataframe, we first just group
+/// all mutations per column, and append them in append-only object.
+///
+/// We represents all of these operations as `ColumnOperation`.
+#[derive(Eq, PartialEq, Debug, Clone, Copy)]
+pub(crate) enum ColumnOperation<T> {
+    NewDoc(DocId),
+    Value(T),
+}
+
+impl<T> From<T> for ColumnOperation<T> {
+    fn from(value: T) -> Self {
+        ColumnOperation::Value(value)
+    }
+}
+
+#[allow(clippy::from_over_into)]
+pub(crate) trait SymbolValue: Into<MiniBuffer> + Clone + Copy + fmt::Debug {
+    fn deserialize(header: NonZeroU8, bytes: &mut &[u8]) -> Result<Self, ParseError>;
+}
+
+pub(crate) struct MiniBuffer {
+    pub bytes: [u8; 9],
+    pub len: usize,
+}
+
+impl MiniBuffer {
+    pub fn as_slice(&self) -> &[u8] {
+        &self.bytes[..self.len]
+    }
+}
+
+fn compute_header_byte(typ: SymbolType, len: usize) -> u8 {
+    assert!(len <= 9);
+    (len << 4) as u8 | typ as u8
+}
+
+impl SymbolValue for NumericalValue {
+    fn deserialize(header_byte: NonZeroU8, bytes: &mut &[u8]) -> Result<Self, ParseError> {
+        let (typ, len) = parse_header_byte(header_byte)?;
+        let value_bytes: &[u8];
+        (value_bytes, *bytes) = bytes.split_at(len);
+        let symbol: NumericalValue = match typ {
+            SymbolType::U64 => {
+                let mut octet: [u8; 8] = [0u8; 8];
+                octet[..value_bytes.len()].copy_from_slice(value_bytes);
+                let val: u64 = u64::from_le_bytes(octet);
+                NumericalValue::U64(val)
+            }
+            SymbolType::I64 => {
+                let mut octet: [u8; 8] = [0u8; 8];
+                octet[..value_bytes.len()].copy_from_slice(value_bytes);
+                let encoded: u64 = u64::from_le_bytes(octet);
+                let val: i64 = decode_zig_zag(encoded);
+                NumericalValue::I64(val)
+            }
+            SymbolType::Float => {
+                let octet: [u8; 8] =
+                    value_bytes.try_into().map_err(|_| ParseError::InvalidLen {
+                        typ: SymbolType::Float,
+                        len,
+                    })?;
+                let val_possibly_nan = f64::from_le_bytes(octet);
+                let val_not_nan = NotNan::new(val_possibly_nan)
+                    .map_err(|_| ParseError::NaN)?;
+                NumericalValue::F64(val_not_nan)
+            }
+        };
+        Ok(symbol)
+    }
+}
+
+#[allow(clippy::from_over_into)]
+impl Into<MiniBuffer> for NumericalValue {
+    fn into(self) -> MiniBuffer {
+        let mut bytes = [0u8; 9];
+        match self {
+            NumericalValue::F64(val) => {
+                let len = 8;
+                let header_byte = compute_header_byte(SymbolType::Float, len);
+                bytes[0] = header_byte;
+                bytes[1..].copy_from_slice(&val.to_le_bytes());
+                MiniBuffer {
+                    bytes,
+                    len: len + 1,
+                }
+            }
+            NumericalValue::U64(val) => {
+                let len = compute_num_bytes_for_u64(val);
+                let header_byte = compute_header_byte(SymbolType::U64, len);
+                bytes[0] = header_byte;
+                bytes[1..].copy_from_slice(&val.to_le_bytes());
+                MiniBuffer {
+                    bytes,
+                    len: len + 1,
+                }
+            }
+            NumericalValue::I64(val) => {
+                let encoded = encode_zig_zag(val);
+                let len = compute_num_bytes_for_u64(encoded);
+                let header_byte = compute_header_byte(SymbolType::I64, len);
+                bytes[0] = header_byte;
+                bytes[1..].copy_from_slice(&encoded.to_le_bytes());
+                MiniBuffer {
+                    bytes,
+                    len: len + 1,
+                }
+            }
+        }
+    }
+}
+
+#[allow(clippy::from_over_into)]
+impl Into<MiniBuffer> for UnorderedId {
+    fn into(self) -> MiniBuffer {
+        let mut bytes = [0u8; 9];
+        let val = self.0 as u64;
+        let len = compute_num_bytes_for_u64(val) + 1;
+        bytes[0] = len as u8;
+        bytes[1..].copy_from_slice(&val.to_le_bytes());
+        MiniBuffer { bytes, len }
+    }
+}
+
+impl SymbolValue for UnorderedId {
+    fn deserialize(header: NonZeroU8, bytes: &mut &[u8]) -> Result<UnorderedId, ParseError> {
+        let len = header.get() as usize;
+        let symbol_bytes: &[u8];
+        (symbol_bytes, *bytes) = bytes.split_at(len);
+        let mut value_bytes = [0u8; 4];
+        value_bytes[..len - 1].copy_from_slice(&symbol_bytes[1..]);
+        let value = u32::from_le_bytes(value_bytes);
+        Ok(UnorderedId(value))
+    }
+}
+
+const HEADER_MASK: u8 = (1u8 << 4) - 1u8;
+
+fn compute_num_bytes_for_u64(val: u64) -> usize {
+    let msb = (64u32 - val.leading_zeros()) as usize;
+    (msb + 7) / 8
+}
+
+fn parse_header_byte(byte: NonZeroU8) -> Result<(SymbolType, usize), ParseError> {
+    let len = (byte.get() as usize) >> 4;
+    let typ_code = byte.get() & HEADER_MASK;
+    let typ = SymbolType::try_from(typ_code)?;
+    Ok((typ, len))
+}
+
+#[derive(Error, Debug)]
+pub enum ParseError {
+    #[error("Type byte unknown `{0}`")]
+    UnknownType(u8),
+    #[error("Invalid len for type `{len}` for type `{typ:?}`.")]
+    InvalidLen { typ: SymbolType, len: usize },
+    #[error("Missing bytes.")]
+    MissingBytes,
+    #[error("Not a number value.")]
+    NaN,
+}
+
+impl<V: SymbolValue> ColumnOperation<V> {
+    pub fn serialize(self) -> MiniBuffer {
+        match self {
+            ColumnOperation::NewDoc(doc) => {
+                let mut minibuf: [u8; 9] = [0u8; 9];
+                minibuf[0] = 0u8;
+                minibuf[1..5].copy_from_slice(&doc.to_le_bytes());
+                MiniBuffer {
+                    bytes: minibuf,
+                    len: 5,
+                }
+            }
+            ColumnOperation::Value(val) => val.into(),
+        }
+    }
+
+    pub fn deserialize(bytes: &mut &[u8]) -> Result<Self, ParseError> {
+        if bytes.is_empty() {
+            return Err(ParseError::MissingBytes);
+        }
+        let header_byte = bytes[0];
+        *bytes = &bytes[1..];
+        if let Some(header_byte) = NonZeroU8::new(header_byte) {
+            let value = V::deserialize(header_byte, bytes)?;
+            Ok(ColumnOperation::Value(value))
+        } else {
+            let doc_bytes: &[u8];
+            (doc_bytes, *bytes) = bytes.split_at(4);
+            let doc: u32 =
+                u32::from_le_bytes(doc_bytes.try_into().map_err(|_| ParseError::MissingBytes)?);
+            Ok(ColumnOperation::NewDoc(doc))
+        }
+    }
+}
+
+#[derive(Copy, Clone, Debug, Eq, PartialEq)]
+#[repr(u8)]
+pub enum SymbolType {
+    U64 = 1u8,
+    I64 = 2u8,
+    Float = 3u8,
+}
+
+impl TryFrom<u8> for SymbolType {
+    type Error = ParseError;
+
+    fn try_from(byte: u8) -> Result<Self, ParseError> {
+        match byte {
+            1u8 => Ok(SymbolType::U64),
+            2u8 => Ok(SymbolType::I64),
+            3u8 => Ok(SymbolType::Float),
+            _ => Err(ParseError::UnknownType(byte)),
+        }
+    }
+}
+
+fn encode_zig_zag(n: i64) -> u64 {
+    ((n << 1) ^ (n >> 63)) as u64
+}
+
+fn decode_zig_zag(n: u64) -> i64 {
+    ((n >> 1) as i64) ^ (-((n & 1) as i64))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{SymbolType, *};
+
+    #[track_caller]
+    fn test_zig_zag_aux(val: i64) {
+        let encoded = super::encode_zig_zag(val);
+        assert_eq!(decode_zig_zag(encoded), val);
+        if let Some(abs_val) = val.checked_abs() {
+            let abs_val = abs_val as u64;
+            assert!(encoded <= abs_val * 2);
+        }
+    }
+
+    #[test]
+    fn test_zig_zag() {
+        assert_eq!(encode_zig_zag(0i64), 0u64);
+        assert_eq!(encode_zig_zag(-1i64), 1u64);
+        assert_eq!(encode_zig_zag(1i64), 2u64);
+        test_zig_zag_aux(0i64);
+        test_zig_zag_aux(i64::MIN);
+        test_zig_zag_aux(i64::MAX);
+    }
+
+    use proptest::prelude::any;
+    use proptest::proptest;
+
+    proptest! {
+        #[test]
+        fn test_proptest_zig_zag(val in any::<i64>()) {
+            test_zig_zag_aux(val);
+        }
+    }
+
+    #[track_caller]
+    fn ser_deser_header_byte_aux(symbol_type: SymbolType, len: usize) {
+        let header_byte = compute_header_byte(symbol_type, len);
+        let (serdeser_numerical_type, serdeser_len) =
+            parse_header_byte(NonZeroU8::new(header_byte).unwrap()).unwrap();
+        assert_eq!(symbol_type, serdeser_numerical_type);
+        assert_eq!(len, serdeser_len);
+    }
+
+    #[test]
+    fn test_header_byte_serialization() {
+        for len in 1..9 {
+            ser_deser_header_byte_aux(SymbolType::Float, len);
+            ser_deser_header_byte_aux(SymbolType::I64, len);
+            ser_deser_header_byte_aux(SymbolType::U64, len);
+        }
+    }
+
+    #[track_caller]
+    fn ser_deser_symbol(symbol: ColumnOperation<NumericalValue>) {
+        let buf = symbol.serialize();
+        let mut bytes = &buf.bytes[..];
+        let serdeser_symbol = ColumnOperation::deserialize(&mut bytes).unwrap();
+        assert_eq!(bytes.len() + buf.len, buf.bytes.len());
+        assert_eq!(symbol, serdeser_symbol);
+    }
+
+    #[test]
+    fn test_compute_num_bytes_for_u64() {
+        assert_eq!(compute_num_bytes_for_u64(0), 0);
+        assert_eq!(compute_num_bytes_for_u64(1), 1);
+        assert_eq!(compute_num_bytes_for_u64(255), 1);
+        assert_eq!(compute_num_bytes_for_u64(256), 2);
+        assert_eq!(compute_num_bytes_for_u64((1 << 16) - 1), 2);
+        assert_eq!(compute_num_bytes_for_u64(1 << 16), 3);
+    }
+
+    #[test]
+    fn test_symbol_serialization() {
+        ser_deser_symbol(ColumnOperation::NewDoc(0));
+        ser_deser_symbol(ColumnOperation::NewDoc(3));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(0i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(1i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(257u64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(-257i64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::I64(i64::MIN)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(0u64)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MIN)));
+        ser_deser_symbol(ColumnOperation::Value(NumericalValue::U64(u64::MAX)));
+    }
+}
diff --git a/columnar/src/writer/mod.rs b/columnar/src/writer/mod.rs
new file mode 100644
index 000000000..8508a6d2e
--- /dev/null
+++ b/columnar/src/writer/mod.rs
@@ -0,0 +1,675 @@
+mod column_operation;
+mod value_index;
+
+use std::io::{self, Write};
+
+use column_operation::ColumnOperation;
+use common::CountingWriter;
+use fastfield_codecs::serialize::ValueIndexInfo;
+use fastfield_codecs::{Column, MonotonicallyMappableToU64, VecColumn};
+use ordered_float::NotNan;
+use stacker::{Addr, ArenaHashMap, ExpUnrolledLinkedList, MemoryArena};
+
+use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality};
+use crate::dictionary::{DictionaryBuilder, IdMapping, UnorderedId};
+use crate::value::{Coerce, NumericalType, NumericalValue};
+use crate::writer::column_operation::SymbolValue;
+use crate::writer::value_index::{IndexBuilder, SpareIndexBuilders};
+use crate::{Cardinality, ColumnarSerializer, DocId};
+
+#[derive(Copy, Clone, Default)]
+struct ColumnWriter {
+    // Detected cardinality of the column so far.
+    cardinality: Cardinality,
+    // Last document inserted.
+    // None if no doc has been added yet.
+    last_doc_opt: Option<u32>,
+    // Buffer containing the serialized values.
+    values: ExpUnrolledLinkedList,
+}
+
+#[derive(Clone, Copy, Default)]
+pub struct NumericalColumnWriter {
+    compatible_numerical_types: CompatibleNumericalTypes,
+    column_writer: ColumnWriter,
+}
+
+#[derive(Clone, Copy)]
+struct CompatibleNumericalTypes {
+    all_values_within_i64_range: bool,
+    all_values_within_u64_range: bool,
+}
+
+impl Default for CompatibleNumericalTypes {
+    fn default() -> CompatibleNumericalTypes {
+        CompatibleNumericalTypes {
+            all_values_within_i64_range: true,
+            all_values_within_u64_range: true,
+        }
+    }
+}
+
+impl CompatibleNumericalTypes {
+    pub fn accept_value(&mut self, numerical_value: NumericalValue) {
+        match numerical_value {
+            NumericalValue::I64(val_i64) => {
+                let value_within_u64_range = val_i64 >= 0i64;
+                self.all_values_within_u64_range &= value_within_u64_range;
+            }
+            NumericalValue::U64(val_u64) => {
+                let value_within_i64_range = val_u64 < i64::MAX as u64;
+                self.all_values_within_i64_range &= value_within_i64_range;
+            }
+            NumericalValue::F64(_) => {
+                self.all_values_within_i64_range = false;
+                self.all_values_within_u64_range = false;
+            }
+        }
+    }
+
+    pub fn to_numerical_type(self) -> NumericalType {
+        if self.all_values_within_i64_range {
+            NumericalType::I64
+        } else if self.all_values_within_u64_range {
+            NumericalType::U64
+        } else {
+            NumericalType::F64
+        }
+    }
+}
+
+impl NumericalColumnWriter {
+    pub fn record_numerical_value(
+        &mut self,
+        doc: DocId,
+        value: NumericalValue,
+        arena: &mut MemoryArena,
+    ) {
+        self.compatible_numerical_types.accept_value(value);
+        self.column_writer.record(doc, value, arena);
+    }
+}
+
+impl ColumnWriter {
+    fn symbol_iterator<'a, V: SymbolValue>(
+        &self,
+        arena: &MemoryArena,
+        buffer: &'a mut Vec<u8>,
+    ) -> impl Iterator<Item = ColumnOperation<V>> + 'a {
+        buffer.clear();
+        self.values.read_to_end(arena, buffer);
+        let mut cursor: &[u8] = &buffer[..];
+        std::iter::from_fn(move || {
+            if cursor.is_empty() {
+                return None;
+            }
+            let symbol = ColumnOperation::deserialize(&mut cursor)
+                .expect("Failed to deserialize symbol from in-memory. This should never happen.");
+            Some(symbol)
+        })
+    }
+
+    fn delta_with_last_doc(&self, doc: DocId) -> u32 {
+        self.last_doc_opt
+            .map(|last_doc| doc - last_doc)
+            .unwrap_or(doc + 1u32)
+    }
+
+    /// Records a change of the document being recorded.
+    ///
+    /// This function will also update the cardinality of the column
+    /// if necessary.
+    fn record(&mut self, doc: DocId, value: NumericalValue, arena: &mut MemoryArena) {
+        // Difference between `doc` and the last doc.
+        match self.delta_with_last_doc(doc) {
+            0 => {
+                // This is the last encounterred document.
+                self.cardinality = Cardinality::Multivalued;
+            }
+            1 => {
+                self.last_doc_opt = Some(doc);
+                self.write_symbol::<NumericalValue>(ColumnOperation::NewDoc(doc), arena);
+            }
+            _ => {
+                self.cardinality = self.cardinality.max(Cardinality::Optional);
+                self.last_doc_opt = Some(doc);
+                self.write_symbol::<NumericalValue>(ColumnOperation::NewDoc(doc), arena);
+            }
+        }
+        self.write_symbol(ColumnOperation::Value(value), arena);
+    }
+
+    // Get the cardinality.
+    // The overall number of docs in the column is necessary to
+    // deal with the case where the all docs contain 1 value, except some documents
+    // at the end of the column.
+    fn get_cardinality(&self, num_docs: DocId) -> Cardinality {
+        if self.delta_with_last_doc(num_docs) > 1 {
+            self.cardinality.max(Cardinality::Optional)
+        } else {
+            self.cardinality
+        }
+    }
+
+    fn write_symbol<V: SymbolValue>(
+        &mut self,
+        symbol: ColumnOperation<V>,
+        arena: &mut MemoryArena,
+    ) {
+        self.values
+            .writer(arena)
+            .extend_from_slice(symbol.serialize().as_slice());
+    }
+}
+
+#[derive(Copy, Clone, Default)]
+pub struct BytesColumnWriter {
+    dictionary_id: u32,
+    column_writer: ColumnWriter,
+}
+
+impl BytesColumnWriter {
+    pub fn with_dictionary_id(dictionary_id: u32) -> BytesColumnWriter {
+        BytesColumnWriter {
+            dictionary_id,
+            column_writer: Default::default(),
+        }
+    }
+
+    pub fn record_bytes(
+        &mut self,
+        doc: DocId,
+        bytes: &[u8],
+        dictionaries: &mut [DictionaryBuilder],
+        arena: &mut MemoryArena,
+    ) {
+        let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
+        let numerical_value = NumericalValue::U64(unordered_id.0 as u64);
+        self.column_writer.record(doc, numerical_value, arena);
+    }
+}
+
+pub struct ColumnarWriter {
+    numerical_field_hash_map: ArenaHashMap,
+    bytes_field_hash_map: ArenaHashMap,
+    arena: MemoryArena,
+    // Dictionaries used to store dictionary-encoded values.
+    dictionaries: Vec<DictionaryBuilder>,
+    buffers: SpareBuffers,
+}
+
+#[derive(Default)]
+struct SpareBuffers {
+    byte_buffer: Vec<u8>,
+    value_index_builders: SpareIndexBuilders,
+    i64_values: Vec<i64>,
+    u64_values: Vec<u64>,
+    f64_values: Vec<ordered_float::NotNan<f64>>,
+}
+
+impl Default for ColumnarWriter {
+    fn default() -> Self {
+        ColumnarWriter {
+            numerical_field_hash_map: ArenaHashMap::new(10_000),
+            bytes_field_hash_map: ArenaHashMap::new(10_000),
+            dictionaries: Vec::new(),
+            arena: MemoryArena::default(),
+            buffers: SpareBuffers::default(),
+        }
+    }
+}
+
+#[derive(Copy, Clone, Ord, PartialOrd, Eq, PartialEq, Debug)]
+enum BytesOrNumerical {
+    Bytes,
+    Numerical,
+}
+
+impl ColumnarWriter {
+    pub fn record_numerical(&mut self, doc: DocId, key: &[u8], numerical_value: NumericalValue) {
+        let (hash_map, arena) = (&mut self.numerical_field_hash_map, &mut self.arena);
+        hash_map.mutate_or_create(key, |column_opt: Option<NumericalColumnWriter>| {
+            let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
+            column.record_numerical_value(doc, numerical_value, arena);
+            column
+        });
+    }
+
+    pub fn record_bytes(&mut self, doc: DocId, key: &[u8], value: &[u8]) {
+        let (hash_map, arena, dictionaries) = (
+            &mut self.bytes_field_hash_map,
+            &mut self.arena,
+            &mut self.dictionaries,
+        );
+        hash_map.mutate_or_create(key, |column_opt: Option<BytesColumnWriter>| {
+            let mut column: BytesColumnWriter = column_opt.unwrap_or_else(|| {
+                let dictionary_id = dictionaries.len() as u32;
+                dictionaries.push(DictionaryBuilder::default());
+                BytesColumnWriter::with_dictionary_id(dictionary_id)
+            });
+            column.record_bytes(doc, value, dictionaries, arena);
+            column
+        });
+    }
+
+    pub fn serialize<W: io::Write>(
+        &mut self,
+        num_docs: DocId,
+        mut serializer: ColumnarSerializer<W>,
+    ) -> io::Result<()> {
+        let mut field_columns: Vec<(&[u8], BytesOrNumerical, Addr)> = self
+            .numerical_field_hash_map
+            .iter()
+            .map(|(term, addr, _)| (term, BytesOrNumerical::Numerical, addr))
+            .collect();
+        field_columns.extend(
+            self.bytes_field_hash_map
+                .iter()
+                .map(|(term, addr, _)| (term, BytesOrNumerical::Bytes, addr)),
+        );
+        let mut key_buffer = Vec::new();
+        field_columns.sort_unstable_by_key(|(key, col_type, _)| (*key, *col_type));
+        let (arena, buffers, dictionaries) = (&self.arena, &mut self.buffers, &self.dictionaries);
+        for (key, bytes_or_numerical, addr) in field_columns {
+            let wrt = serializer.wrt();
+            let start_offset = wrt.written_bytes();
+            let column_type_and_cardinality: ColumnTypeAndCardinality =
+                match bytes_or_numerical {
+                BytesOrNumerical::Bytes => {
+                    let BytesColumnWriter { dictionary_id, column_writer } =
+                        self.bytes_field_hash_map.read(addr);
+                    let dictionary_builder =
+                        &dictionaries[dictionary_id as usize];
+                    serialize_bytes_column(
+                        &column_writer,
+                        num_docs,
+                        dictionary_builder,
+                        arena,
+                        buffers,
+                        wrt,
+                    )?;
+                    ColumnTypeAndCardinality {
+                        cardinality: column_writer.get_cardinality(num_docs),
+                        typ: ColumnType::Bytes,
+                    }
+                }
+                BytesOrNumerical::Numerical => {
+                    let NumericalColumnWriter { compatible_numerical_types, column_writer  } =
+                        self.numerical_field_hash_map.read(addr);
+                    let cardinality = column_writer.get_cardinality(num_docs);
+                    let numerical_type = compatible_numerical_types.to_numerical_type();
+                    serialize_numerical_column(
+                        cardinality,
+                        numerical_type,
+                        &column_writer,
+                        num_docs,
+                        arena,
+                        buffers,
+                        wrt,
+                    )?;
+                    ColumnTypeAndCardinality {
+                        cardinality,
+                        typ: ColumnType::Numerical(numerical_type),
+                    }
+                }
+            };
+            let end_offset = wrt.written_bytes();
+            let key_with_type = prepare_key(key, column_type_and_cardinality, &mut key_buffer);
+            serializer.record_column_offsets(key_with_type, start_offset..end_offset)?;
+        }
+        serializer.finalize()?;
+        Ok(())
+    }
+}
+
+/// Returns a key consisting of the concatenation of the key and the column_type_and_cardinality
+/// code.
+fn prepare_key<'a>(
+    key: &[u8],
+    column_type_cardinality: ColumnTypeAndCardinality,
+    buffer: &'a mut Vec<u8>,
+) -> &'a [u8] {
+    buffer.clear();
+    buffer.extend_from_slice(key);
+    buffer.push(0u8);
+    buffer.push(column_type_cardinality.to_code());
+    &buffer[..]
+}
+
+fn serialize_bytes_column<W: io::Write>(
+    column_writer: &ColumnWriter,
+    num_docs: DocId,
+    dictionary_builder: &DictionaryBuilder,
+    arena: &MemoryArena,
+    buffers: &mut SpareBuffers,
+    wrt: &mut CountingWriter<W>,
+) -> io::Result<()> {
+    let start_offset = wrt.written_bytes();
+    let id_mapping: IdMapping = dictionary_builder.serialize(wrt)?;
+    let dictionary_num_bytes: u32 = (wrt.written_bytes() - start_offset) as u32;
+    let cardinality = column_writer.get_cardinality(num_docs);
+    let SpareBuffers {
+        byte_buffer,
+        value_index_builders,
+        u64_values,
+        ..
+    } = buffers;
+    let symbol_iterator = column_writer
+        .symbol_iterator(arena, byte_buffer)
+        .map(|symbol: ColumnOperation<UnorderedId>| {
+            // We map unordered ids to ordered ids.
+            match symbol {
+                ColumnOperation::Value(unordered_id) => {
+                    let ordered_id = id_mapping.to_ord(unordered_id);
+                    ColumnOperation::Value(ordered_id.0 as u64)
+                }
+                ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
+            }
+        });
+    serialize_column(
+        symbol_iterator,
+        cardinality,
+        num_docs,
+        value_index_builders,
+        u64_values,
+        wrt,
+    )?;
+    wrt.write_all(&dictionary_num_bytes.to_le_bytes()[..])?;
+    Ok(())
+}
+
+fn serialize_numerical_column<W: io::Write>(
+    cardinality: Cardinality,
+    numerical_type: NumericalType,
+    column_writer: &ColumnWriter,
+    num_docs: DocId,
+    arena: &MemoryArena,
+    buffers: &mut SpareBuffers,
+    wrt: &mut W,
+) -> io::Result<()> {
+    let SpareBuffers {
+        byte_buffer,
+        value_index_builders,
+        u64_values,
+        i64_values,
+        f64_values,
+    } = buffers;
+    let symbol_iterator = column_writer.symbol_iterator(arena, byte_buffer);
+    match numerical_type {
+        NumericalType::I64 => {
+            serialize_column(
+                coerce_numerical_symbol::<i64>(symbol_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                i64_values,
+                wrt,
+            )?;
+        }
+        NumericalType::U64 => {
+            serialize_column(
+                coerce_numerical_symbol::<u64>(symbol_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                u64_values,
+                wrt,
+            )?;
+        }
+        NumericalType::F64 => {
+            serialize_column(
+                coerce_numerical_symbol::<NotNan<f64>>(symbol_iterator),
+                cardinality,
+                num_docs,
+                value_index_builders,
+                f64_values,
+                wrt,
+            )?;
+        }
+    };
+    Ok(())
+}
+
+fn serialize_column<
+    T: Copy + Ord + Default + Send + Sync + MonotonicallyMappableToU64,
+    W: io::Write,
+>(
+    symbol_iterator: impl Iterator<Item = ColumnOperation<T>>,
+    cardinality: Cardinality,
+    num_docs: DocId,
+    value_index_builders: &mut SpareIndexBuilders,
+    values: &mut Vec<T>,
+    wrt: &mut W,
+) -> io::Result<()>
+where
+    for<'a> VecColumn<'a, T>: Column<T>,
+{
+    match cardinality {
+        Cardinality::Required => {
+            consume_symbol_iterator(
+                symbol_iterator,
+                value_index_builders.borrow_required_index_builder(),
+                values,
+            );
+            fastfield_codecs::serialize(
+                VecColumn::from(&values[..]),
+                wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+        Cardinality::Optional => {
+            let optional_index_builder = value_index_builders.borrow_optional_index_builder();
+            consume_symbol_iterator(symbol_iterator, optional_index_builder, values);
+            let optional_index = optional_index_builder.finish(num_docs);
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::SingleValue(Box::new(optional_index)),
+                VecColumn::from(&values[..]),
+                wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+        Cardinality::Multivalued => {
+            let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
+            consume_symbol_iterator(symbol_iterator, multivalued_index_builder, values);
+            let multivalued_index = multivalued_index_builder.finish(num_docs);
+            fastfield_codecs::serialize::serialize_new(
+                ValueIndexInfo::MultiValue(Box::new(multivalued_index)),
+                VecColumn::from(&values[..]),
+                wrt,
+                &fastfield_codecs::ALL_CODEC_TYPES[..],
+            )?;
+        }
+    }
+    Ok(())
+}
+
+fn coerce_numerical_symbol<T>(
+    symbol_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
+) -> impl Iterator<Item = ColumnOperation<T>>
+where T: Coerce {
+    symbol_iterator.map(|symbol| match symbol {
+        ColumnOperation::NewDoc(doc) => ColumnOperation::NewDoc(doc),
+        ColumnOperation::Value(numerical_value) => {
+            ColumnOperation::Value(Coerce::coerce(numerical_value))
+        }
+    })
+}
+
+fn consume_symbol_iterator<T, TIndexBuilder: IndexBuilder>(
+    symbol_iterator: impl Iterator<Item = ColumnOperation<T>>,
+    index_builder: &mut TIndexBuilder,
+    values: &mut Vec<T>,
+) {
+    for symbol in symbol_iterator {
+        match symbol {
+            ColumnOperation::NewDoc(doc) => {
+                index_builder.record_doc(doc);
+            }
+            ColumnOperation::Value(value) => {
+                index_builder.record_value();
+                values.push(value);
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use ordered_float::NotNan;
+    use stacker::MemoryArena;
+
+    use super::prepare_key;
+    use crate::column_type_header::{ColumnType, ColumnTypeAndCardinality};
+    use crate::value::{NumericalType, NumericalValue};
+    use crate::writer::column_operation::ColumnOperation;
+    use crate::writer::CompatibleNumericalTypes;
+    use crate::Cardinality;
+
+    #[test]
+    fn test_prepare_key_bytes() {
+        let mut buffer: Vec<u8> = b"somegarbage".to_vec();
+        let column_type_and_cardinality = ColumnTypeAndCardinality {
+            typ: ColumnType::Bytes,
+            cardinality: Cardinality::Optional,
+        };
+        let prepared_key = prepare_key(b"root\0child", column_type_and_cardinality, &mut buffer);
+        assert_eq!(prepared_key.len(), 12);
+        assert_eq!(&prepared_key[..10], b"root\0child");
+        assert_eq!(prepared_key[10], 0u8);
+        assert_eq!(prepared_key[11], column_type_and_cardinality.to_code());
+    }
+
+    #[test]
+    fn test_column_writer_required_simple() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, 14i64.into(), &mut arena);
+        column_writer.record(1u32, 15i64.into(), &mut arena);
+        column_writer.record(2u32, (-16i64).into(), &mut arena);
+        assert_eq!(column_writer.get_cardinality(3), Cardinality::Required);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .symbol_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 6);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(14i64))
+        ));
+        assert!(matches!(symbols[2], ColumnOperation::NewDoc(1u32)));
+        assert!(matches!(
+            symbols[3],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+        assert!(matches!(symbols[4], ColumnOperation::NewDoc(2u32)));
+        assert!(matches!(
+            symbols[5],
+            ColumnOperation::Value(NumericalValue::I64(-16i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_optional_cardinality_missing_first() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(1u32, 15i64.into(), &mut arena);
+        column_writer.record(2u32, (-16i64).into(), &mut arena);
+        assert_eq!(column_writer.get_cardinality(3), Cardinality::Optional);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .symbol_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 4);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(1u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+        assert!(matches!(symbols[2], ColumnOperation::NewDoc(2u32)));
+        assert!(matches!(
+            symbols[3],
+            ColumnOperation::Value(NumericalValue::I64(-16i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_optional_cardinality_missing_last() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, 15i64.into(), &mut arena);
+        assert_eq!(column_writer.get_cardinality(2), Cardinality::Optional);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .symbol_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 2);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(15i64))
+        ));
+    }
+
+    #[test]
+    fn test_column_writer_multivalued() {
+        let mut arena = MemoryArena::default();
+        let mut column_writer = super::ColumnWriter::default();
+        column_writer.record(0u32, 16i64.into(), &mut arena);
+        column_writer.record(0u32, 17i64.into(), &mut arena);
+        assert_eq!(column_writer.get_cardinality(1), Cardinality::Multivalued);
+        let mut buffer = Vec::new();
+        let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
+            .symbol_iterator(&mut arena, &mut buffer)
+            .collect();
+        assert_eq!(symbols.len(), 3);
+        assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
+        assert!(matches!(
+            symbols[1],
+            ColumnOperation::Value(NumericalValue::I64(16i64))
+        ));
+        assert!(matches!(
+            symbols[2],
+            ColumnOperation::Value(NumericalValue::I64(17i64))
+        ));
+    }
+
+    #[track_caller]
+    fn test_column_writer_coercion_iter_aux(
+        values: impl Iterator<Item = NumericalValue>,
+        expected_numerical_type: NumericalType,
+    ) {
+        let mut compatible_numerical_types = CompatibleNumericalTypes::default();
+        for value in values {
+            compatible_numerical_types.accept_value(value);
+        }
+        assert_eq!(
+            compatible_numerical_types.to_numerical_type(),
+            expected_numerical_type
+        );
+    }
+
+    #[track_caller]
+    fn test_column_writer_coercion_aux(
+        values: &[NumericalValue],
+        expected_numerical_type: NumericalType,
+    ) {
+        test_column_writer_coercion_iter_aux(values.iter().copied(), expected_numerical_type);
+        test_column_writer_coercion_iter_aux(values.iter().rev().copied(), expected_numerical_type);
+    }
+
+    #[test]
+    fn test_column_writer_coercion() {
+        test_column_writer_coercion_aux(&[], NumericalType::I64);
+        test_column_writer_coercion_aux(&[1i64.into()], NumericalType::I64);
+        test_column_writer_coercion_aux(&[1u64.into()], NumericalType::I64);
+        // We don't detect exact integer at the moment. We could!
+        test_column_writer_coercion_aux(&[NotNan::new(1f64).unwrap().into()], NumericalType::F64);
+        test_column_writer_coercion_aux(&[u64::MAX.into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[(i64::MAX as u64).into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[(1u64 << 63).into()], NumericalType::U64);
+        test_column_writer_coercion_aux(&[1i64.into(), 1u64.into()], NumericalType::I64);
+        test_column_writer_coercion_aux(&[u64::MAX.into(), (-1i64).into()], NumericalType::F64);
+    }
+}
diff --git a/columnar/src/writer/value_index.rs b/columnar/src/writer/value_index.rs
new file mode 100644
index 000000000..7284eb845
--- /dev/null
+++ b/columnar/src/writer/value_index.rs
@@ -0,0 +1,218 @@
+use fastfield_codecs::serialize::{MultiValueIndexInfo, SingleValueIndexInfo};
+
+use crate::DocId;
+
+/// The `IndexBuilder` interprets a sequence of
+/// calls of the form:
+/// (record_doc,record_value+)*
+/// and can then serialize the results into an index.
+///
+/// It has different implementation depending on whether the
+/// cardinality is required, optional, or multivalued.
+pub(crate) trait IndexBuilder {
+    fn record_doc(&mut self, doc: DocId);
+    #[inline]
+    fn record_value(&mut self) {}
+}
+
+/// The RequiredIndexBuilder does nothing.
+#[derive(Default)]
+pub struct RequiredIndexBuilder;
+
+impl IndexBuilder for RequiredIndexBuilder {
+    #[inline(always)]
+    fn record_doc(&mut self, _doc: DocId) {}
+}
+
+#[derive(Default)]
+pub struct OptionalIndexBuilder {
+    docs: Vec<DocId>,
+}
+
+struct SingleValueArrayIndex<'a> {
+    docs: &'a [DocId],
+    num_docs: DocId,
+}
+
+impl<'a> SingleValueIndexInfo for SingleValueArrayIndex<'a> {
+    fn num_vals(&self) -> u32 {
+        self.num_docs as u32
+    }
+
+    fn num_non_nulls(&self) -> u32 {
+        self.docs.len() as u32
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        Box::new(self.docs.iter().copied())
+    }
+}
+
+impl OptionalIndexBuilder {
+    pub fn finish(&mut self, num_docs: DocId) -> impl SingleValueIndexInfo + '_ {
+        debug_assert!(self
+            .docs
+            .last()
+            .copied()
+            .map(|last_doc| last_doc < num_docs)
+            .unwrap_or(true));
+        SingleValueArrayIndex {
+            docs: &self.docs[..],
+            num_docs,
+        }
+    }
+
+    fn reset(&mut self) {
+        self.docs.clear();
+    }
+}
+
+impl IndexBuilder for OptionalIndexBuilder {
+    #[inline(always)]
+    fn record_doc(&mut self, doc: DocId) {
+        debug_assert!(self
+            .docs
+            .last()
+            .copied()
+            .map(|prev_doc| doc > prev_doc)
+            .unwrap_or(true));
+        self.docs.push(doc);
+    }
+}
+
+#[derive(Default)]
+pub struct MultivaluedIndexBuilder {
+    // TODO should we switch to `start_offset`?
+    end_values: Vec<DocId>,
+    total_num_vals_seen: u32,
+}
+
+pub struct MultivaluedValueArrayIndex<'a> {
+    end_offsets: &'a [DocId],
+}
+
+impl<'a> MultiValueIndexInfo for MultivaluedValueArrayIndex<'a> {
+    fn num_docs(&self) -> u32 {
+        self.end_offsets.len() as u32
+    }
+
+    fn num_vals(&self) -> u32 {
+        self.end_offsets.last().copied().unwrap_or(0u32)
+    }
+
+    fn iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
+        if self.end_offsets.is_empty() {
+            return Box::new(std::iter::empty());
+        }
+        let n = self.end_offsets.len();
+        Box::new(std::iter::once(0u32).chain(self.end_offsets[..n - 1].iter().copied()))
+    }
+}
+
+impl MultivaluedIndexBuilder {
+    pub fn finish(&mut self, num_docs: DocId) -> impl MultiValueIndexInfo + '_ {
+        self.end_values
+            .resize(num_docs as usize, self.total_num_vals_seen);
+        MultivaluedValueArrayIndex {
+            end_offsets: &self.end_values[..],
+        }
+    }
+
+    fn reset(&mut self) {
+        self.end_values.clear();
+        self.total_num_vals_seen = 0;
+    }
+}
+
+impl IndexBuilder for MultivaluedIndexBuilder {
+    fn record_doc(&mut self, doc: DocId) {
+        self.end_values
+            .resize(doc as usize, self.total_num_vals_seen);
+    }
+
+    fn record_value(&mut self) {
+        self.total_num_vals_seen += 1;
+    }
+}
+
+/// The `SpareIndexBuilders` is there to avoid allocating a
+/// new index builder for every single column.
+#[derive(Default)]
+pub struct SpareIndexBuilders {
+    required_index_builder: RequiredIndexBuilder,
+    optional_index_builder: OptionalIndexBuilder,
+    multivalued_index_builder: MultivaluedIndexBuilder,
+}
+
+impl SpareIndexBuilders {
+    pub fn borrow_required_index_builder(&mut self) -> &mut RequiredIndexBuilder {
+        &mut self.required_index_builder
+    }
+
+    pub fn borrow_optional_index_builder(&mut self) -> &mut OptionalIndexBuilder {
+        self.optional_index_builder.reset();
+        &mut self.optional_index_builder
+    }
+
+    pub fn borrow_multivalued_index_builder(&mut self) -> &mut MultivaluedIndexBuilder {
+        self.multivalued_index_builder.reset();
+        &mut self.multivalued_index_builder
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_optional_value_index_builder() {
+        let mut opt_value_index_builder = OptionalIndexBuilder::default();
+        opt_value_index_builder.record_doc(0u32);
+        opt_value_index_builder.record_value();
+        assert_eq!(
+            &opt_value_index_builder
+                .finish(1u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            &[0]
+        );
+        opt_value_index_builder.reset();
+        opt_value_index_builder.record_doc(1u32);
+        opt_value_index_builder.record_value();
+        assert_eq!(
+            &opt_value_index_builder
+                .finish(2u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            &[1]
+        );
+    }
+
+    #[test]
+    fn test_multivalued_value_index_builder() {
+        let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
+        multivalued_value_index_builder.record_doc(1u32);
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_doc(2u32);
+        multivalued_value_index_builder.record_value();
+        assert_eq!(
+            multivalued_value_index_builder
+                .finish(4u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            vec![0, 0, 2, 3]
+        );
+        multivalued_value_index_builder.reset();
+        multivalued_value_index_builder.record_doc(2u32);
+        multivalued_value_index_builder.record_value();
+        multivalued_value_index_builder.record_value();
+        assert_eq!(
+            multivalued_value_index_builder
+                .finish(4u32)
+                .iter()
+                .collect::<Vec<u32>>(),
+            vec![0, 0, 0, 2]
+        );
+    }
+}
diff --git a/fastfield_codecs/src/lib.rs b/fastfield_codecs/src/lib.rs
index 1b4a2d6ea..6c8e9e267 100644
--- a/fastfield_codecs/src/lib.rs
+++ b/fastfield_codecs/src/lib.rs
@@ -43,7 +43,7 @@ mod null_index_footer;
 
 mod column;
 mod gcd;
-mod serialize;
+pub mod serialize;
 
 pub use ordered_float;