feat: replace fjall with bichon-blob for blob storage

- use a single Engine instance for email + attachment blobs
  - add delete_batch, gc_if_needed, background flush to blob crate
  - fix Entry.raw_size storing compressed length instead of original
  - move fjall-dependent migration code from core to admin crate
  - add STORAGE_VERSION file for layout version detection
This commit is contained in:
rustmailer
2026-07-10 03:17:38 +08:00
parent 4cdf3ee5f1
commit 36692e2091
14 changed files with 396 additions and 381 deletions
+69
View File
@@ -0,0 +1,69 @@
//
// Copyright (c) 2025 rustmailer.com (https://rustmailer.com)
//
// This file is part of the Bichon Email Archiving Project
//
// 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/>.
use tantivy::schema::Field;
pub const F_MESSAGE_ID: &str = "message_id";
pub const F_ACCOUNT_ID: &str = "account_id";
pub const F_MAILBOX_ID: &str = "mailbox_id";
pub const F_UID: &str = "uid";
pub const F_SUBJECT: &str = "subject";
pub const F_TEXT: &str = "text";
pub const F_FROM: &str = "from";
pub const F_TO: &str = "to";
pub const F_CC: &str = "cc";
pub const F_BCC: &str = "bcc";
pub const F_DATE: &str = "date";
pub const F_INTERNAL_DATE: &str = "internal_date";
pub const F_SIZE: &str = "size";
pub const F_THREAD_ID: &str = "thread_id";
pub const F_ATTACHMENTS: &str = "attachments";
pub const F_HAS_ATTACHMENT: &str = "has_attachment";
pub const F_TAGS: &str = "tags";
pub const F_ID: &str = "id";
pub struct EnvelopeFields {
pub f_id: Field,
pub f_message_id: Field,
pub f_account_id: Field,
pub f_mailbox_id: Field,
pub f_uid: Field,
pub f_subject: Field,
pub f_text: Field,
pub f_from: Field,
pub f_to: Field,
pub f_cc: Field,
pub f_bcc: Field,
pub f_date: Field,
pub f_internal_date: Field,
pub f_size: Field,
pub f_thread_id: Field,
pub f_attachments: Field,
pub f_has_attachment: Field,
pub f_tags: Field,
}
pub const F_EML: &str = "eml";
pub struct EmlFields {
pub f_id: Field,
pub f_account_id: Field,
pub f_mailbox_id: Field,
pub f_eml: Field,
}
+2
View File
@@ -0,0 +1,2 @@
pub mod fields;
pub mod schema;
+114
View File
@@ -0,0 +1,114 @@
use tantivy::schema::{FacetOptions, Field, Schema, FAST, INDEXED, STORED, STRING, TEXT};
use crate::legacy::fields::{EmlFields, EnvelopeFields, *};
pub struct SchemaTools;
impl SchemaTools {
pub fn envelope_schema() -> Schema {
EnvelopeSchema::build().0
}
pub fn eml_schema() -> Schema {
EmlSchema::build().0
}
pub fn envelope_fields() -> EnvelopeFields {
EnvelopeSchema::fields()
}
pub fn eml_fields() -> EmlFields {
EmlSchema::fields()
}
pub fn envelope_default_fields() -> Vec<Field> {
let f = Self::envelope_fields();
vec![f.f_subject, f.f_text, f.f_attachments]
}
}
// ─── Schema builders ──────────────────────────────────────────────────────────
struct EnvelopeSchema;
impl EnvelopeSchema {
fn build() -> (Schema, EnvelopeFields) {
let mut b = Schema::builder();
let f_id = b.add_u64_field(F_ID, INDEXED | STORED | FAST);
let f_account_id = b.add_u64_field(F_ACCOUNT_ID, INDEXED | STORED | FAST);
let f_mailbox_id = b.add_u64_field(F_MAILBOX_ID, INDEXED | STORED | FAST);
let f_uid = b.add_u64_field(F_UID, INDEXED | STORED | FAST);
let f_thread_id = b.add_u64_field(F_THREAD_ID, INDEXED | STORED | FAST);
let f_subject = b.add_text_field(F_SUBJECT, TEXT | STORED);
let f_text = b.add_text_field(F_TEXT, TEXT | STORED);
let f_attachments = b.add_text_field(F_ATTACHMENTS, TEXT | STORED);
let f_from = b.add_text_field(F_FROM, STRING | STORED | FAST);
let f_to = b.add_text_field(F_TO, STRING | STORED);
let f_cc = b.add_text_field(F_CC, STRING | STORED);
let f_bcc = b.add_text_field(F_BCC, STRING | STORED);
let f_message_id = b.add_text_field(F_MESSAGE_ID, STRING | STORED);
let f_date = b.add_i64_field(F_DATE, STORED | FAST);
let f_internal_date = b.add_i64_field(F_INTERNAL_DATE, STORED | FAST);
let f_size = b.add_u64_field(F_SIZE, STORED | FAST);
let f_has_attachment = b.add_bool_field(F_HAS_ATTACHMENT, INDEXED | STORED | FAST);
let f_tags = b.add_facet_field(F_TAGS, FacetOptions::default().set_stored());
let fields = EnvelopeFields {
f_id,
f_account_id,
f_mailbox_id,
f_uid,
f_thread_id,
f_subject,
f_text,
f_attachments,
f_from,
f_to,
f_cc,
f_bcc,
f_message_id,
f_date,
f_internal_date,
f_size,
f_has_attachment,
f_tags,
};
(b.build(), fields)
}
fn fields() -> EnvelopeFields {
Self::build().1
}
}
struct EmlSchema;
impl EmlSchema {
fn build() -> (Schema, EmlFields) {
let mut b = Schema::builder();
let f_id = b.add_u64_field(F_ID, INDEXED | FAST);
let f_account_id = b.add_u64_field(F_ACCOUNT_ID, INDEXED | STORED | FAST);
let f_mailbox_id = b.add_u64_field(F_MAILBOX_ID, INDEXED | STORED | FAST);
let f_eml = b.add_bytes_field(F_EML, STORED);
let fields = EmlFields {
f_id,
f_account_id,
f_mailbox_id,
f_eml,
};
(b.build(), fields)
}
fn fields() -> EmlFields {
Self::build().1
}
}
+2
View File
@@ -21,8 +21,10 @@ use dialoguer::{theme::ColorfulTheme, Select};
use crate::{migrate::handle_migration, reset::handle_reset_password};
pub mod legacy;
pub mod meta;
pub mod migrate;
pub mod migrate_store;
pub mod reset;
+228 -10
View File
@@ -1,12 +1,23 @@
use std::path::{Path, PathBuf};
use std::{collections::HashMap, path::PathBuf};
use bichon_core::migrate::{
count_eml_segments, do_migrate_segment, is_tantivy_index_dir,
store::{LegacyDirs, NewDirs, NewIndexWriter},
use bichon_core::{
error::{code::ErrorCode, BichonResult},
migrate::is_tantivy_index_dir,
raise_error,
};
use console::style;
use dialoguer::{theme::ColorfulTheme, Confirm, Input};
use indicatif::{ProgressBar, ProgressStyle};
use tantivy::{
collector::TopDocs,
columnar::Column,
query::TermQuery,
schema::{IndexRecordOption, Value},
DocAddress, Index, TantivyDocument, Term,
};
use crate::legacy::schema::SchemaTools;
use crate::migrate_store::{LegacyDirs, NewDirs, NewIndexWriter};
pub fn handle_migration(theme: &ColorfulTheme) {
println!(
@@ -21,7 +32,7 @@ pub fn handle_migration(theme: &ColorfulTheme) {
style(
"This tool migrates data from the legacy v0.3.7 Tantivy-based storage \
architecture to the new v1.x \
separated index and Fjall-backed storage format."
separated index and blob-backed storage format."
)
.dim()
);
@@ -35,8 +46,8 @@ pub fn handle_migration(theme: &ColorfulTheme) {
New v1.x architecture:\n\
• mail indexes stored in Tantivy\n\
• attachment indexes stored in Tantivy\n\
• raw message data stored in Fjall\n\
• attachment blobs stored in Fjall"
• raw message data stored in blob engine\n\
• attachment blobs stored in blob engine"
)
.dim()
);
@@ -54,7 +65,7 @@ pub fn handle_migration(theme: &ColorfulTheme) {
let root_dir_str: String = Input::with_theme(theme)
.with_prompt("Enter --bichon-root-dir (same value used by the old server)")
.validate_with(|input: &String| -> Result<(), &str> {
let path = Path::new(input);
let path = PathBuf::from(input);
if !path.is_absolute() {
return Err("Path must be absolute.");
}
@@ -81,7 +92,7 @@ pub fn handle_migration(theme: &ColorfulTheme) {
if input.is_empty() {
return Ok(());
}
let path = Path::new(input);
let path = PathBuf::from(input);
if !path.is_absolute() {
return Err("Path must be absolute.");
}
@@ -119,7 +130,7 @@ pub fn handle_migration(theme: &ColorfulTheme) {
if input.is_empty() {
return Ok(());
}
let path = Path::new(input);
let path = PathBuf::from(input);
if !path.is_absolute() {
return Err("Path must be absolute.");
}
@@ -447,3 +458,210 @@ pub fn is_legacy_data_layout_with_paths(
Ok(envelope_result || eml_result)
}
/// Return the number of segments in the legacy EML Tantivy index.
/// Each segment can be passed to `do_migrate_segment` for bounded-memory batch migration.
pub fn count_eml_segments(legacy: &LegacyDirs) -> BichonResult<usize> {
let eml_index = Index::open_in_dir(&legacy.eml_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let reader = eml_index
.reader()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let searcher = reader.searcher();
Ok(searcher.segment_readers().len())
}
/// Migrate all documents from a single EML segment to the new storage layout.
///
/// This is the core of the batch migration strategy: each Process B invocation
/// handles exactly one EML segment, so peak memory is bounded by that segment's
/// size regardless of the total archive size.
pub fn do_migrate_segment<F>(
batch_size: u32,
legacy: LegacyDirs,
writer: &mut NewIndexWriter,
segment_index: usize,
mut on_progress: F,
) -> BichonResult<()>
where
F: FnMut(&str),
{
// ── open legacy indices ────────────────────────────────────────────
let envelope_index = Index::open_in_dir(&legacy.envelope_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let eml_index = Index::open_in_dir(&legacy.eml_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let envelope_reader = envelope_index
.reader()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let eml_reader = eml_index
.reader()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let envelope_searcher = envelope_reader.searcher();
let eml_searcher = eml_reader.searcher();
let ef = SchemaTools::envelope_fields();
let mf = SchemaTools::eml_fields();
let eml_segments = eml_searcher.segment_readers();
let eml_segment = eml_segments.get(segment_index).ok_or_else(|| {
raise_error!(
format!(
"segment index {} out of range ({} segments)",
segment_index,
eml_segments.len()
),
ErrorCode::InternalError
)
})?;
let num_docs = eml_segment.num_docs();
if num_docs == 0 {
on_progress("TOTAL:0");
on_progress("DONE:0:0");
return Ok(());
}
on_progress(&format!("TOTAL:{}", num_docs));
let max_doc = eml_segment.max_doc();
let ff = eml_segment.fast_fields();
let f_id_col: Column<u64> = ff.u64("id").map_err(|e| {
raise_error!(
format!("failed to open f_id fast field: {e:#?}"),
ErrorCode::InternalError
)
})?;
// ── Phase 1: build eid → (uid, internal_date) from envelope, then drop it ──
let mut envelope_map: HashMap<u64, (u32, i64)> = HashMap::with_capacity(num_docs as usize);
let mut env_scanned = 0u32;
let mut env_skipped = 0u32;
for doc_id in 0..max_doc {
if eml_segment.is_deleted(doc_id) {
continue;
}
let eid = f_id_col.values.get_val(doc_id);
let term = Term::from_field_u64(ef.f_id, eid);
let query = TermQuery::new(term, IndexRecordOption::Basic);
let hits: Vec<(_, DocAddress)> = envelope_searcher
.search(&query, &TopDocs::with_limit(1).order_by_score())
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
if let Some((_, addr)) = hits.first() {
let env_doc: TantivyDocument = envelope_searcher
.doc(*addr)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let uid = env_doc
.get_first(ef.f_uid)
.and_then(|v| v.as_u64())
.unwrap_or(0) as u32;
let internal_date = env_doc
.get_first(ef.f_internal_date)
.and_then(|v| v.as_i64())
.unwrap_or(0);
envelope_map.insert(eid, (uid, internal_date));
env_scanned += 1;
} else {
env_skipped += 1;
}
if env_scanned % 10 == 0 {
on_progress(&format!(
"PHASE1:{}/{} skipped:{}",
env_scanned, max_doc, env_skipped
));
}
}
// Free the envelope index before the heavy EML processing.
drop(envelope_searcher);
drop(envelope_reader);
drop(envelope_index);
// ── Phase 2: process EML docs, streaming one at a time ─────────────
let mut total_migrated = 0usize;
let mut total_skipped = 0usize;
let mut chunk_start = 0u32;
while chunk_start < max_doc {
let chunk_end = (chunk_start + batch_size).min(max_doc);
let store_reader = eml_segment
.get_store_reader(2)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
for doc_id in chunk_start..chunk_end {
if eml_segment.is_deleted(doc_id) {
continue;
}
let eid = f_id_col.values.get_val(doc_id);
let (uid, internal_date) = match envelope_map.get(&eid) {
Some(v) => *v,
None => {
on_progress(&format!("WARN: eid {} envelope not found", eid));
total_skipped += 1;
continue;
}
};
let eml_doc: TantivyDocument = store_reader
.get(doc_id)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let account_id = match eml_doc.get_first(mf.f_account_id).and_then(|v| v.as_u64()) {
Some(v) => v,
None => {
on_progress(&format!("WARN: eid {} account_id missing", eid));
total_skipped += 1;
continue;
}
};
let mailbox_id = eml_doc
.get_first(mf.f_mailbox_id)
.and_then(|v| v.as_u64())
.unwrap_or(0);
// Borrow directly from eml_doc — no .to_vec() clone.
let eml_bytes = match eml_doc.get_first(mf.f_eml).and_then(|v| v.as_bytes()) {
Some(b) => b,
None => {
on_progress(&format!("WARN: eid {} eml bytes missing", eid));
total_skipped += 1;
continue;
}
};
if let Err(e) = writer.ingest(eml_bytes, account_id, mailbox_id, uid, internal_date) {
on_progress(&format!(
"ERROR: Account {} eid {} ingest failed: {}",
account_id, eid, e
));
total_skipped += 1;
continue;
}
total_migrated += 1;
if total_migrated % 10 == 0 || total_migrated as u32 == num_docs {
on_progress(&format!("PROGRESS:{}:{}", total_migrated, num_docs));
}
}
drop(store_reader);
// Flush blob buffers via ingestion API — bypasses memtable/WAL.
writer.flush_fjall_buffers()?;
chunk_start = chunk_end;
}
on_progress(&format!("DONE:{}:{}", total_migrated, total_skipped));
Ok(())
}
+560
View File
@@ -0,0 +1,560 @@
use std::{path::PathBuf, time::Instant};
use bytes::Bytes;
use mail_parser::MimeHeaders;
use bichon_core::{
envelope::extractor::extract_references, message::content::AttachmentInfo,
store::tantivy::tokenizers::EuroTokenizer, utils::compute_content_hash,
};
use fjall::{
config::{BlockSizePolicy, CompressionPolicy},
CompressionType, Database, Keyspace, KeyspaceCreateOptions, KvSeparationOptions,
};
use mail_parser::MessageParser;
use tantivy::{indexer::NoMergePolicy, Index, IndexWriter, TantivyDocument};
use uuid::Uuid;
use bichon_core::{
common::AddrVec,
envelope::extractor::{compute_thread_id, generate_message_id},
error::{code::ErrorCode, BichonResult},
raise_error,
store::envelope::Envelope,
store::tantivy::{
model::{AttachmentModel, EnvelopeWithAttachments},
schema::SchemaTools,
},
utc_now,
};
pub struct LegacyDirs {
pub envelope_dir: PathBuf,
pub eml_dir: PathBuf,
}
pub struct NewDirs {
pub envelope_dir: PathBuf,
pub attachment_dir: PathBuf,
pub storage_dir: PathBuf,
}
impl LegacyDirs {
pub fn new(index: PathBuf, data: PathBuf) -> Self {
Self {
envelope_dir: index,
eml_dir: data,
}
}
}
impl NewDirs {
pub fn new(index: PathBuf, data: PathBuf) -> Self {
Self {
envelope_dir: index.join("mail_metadata"),
attachment_dir: index.join("attachment_metadata"),
storage_dir: data,
}
}
}
pub struct DetachOutput {
pub infos: Vec<AttachmentInfo>,
pub blobs: Vec<(String, Bytes)>,
}
pub fn detach_attachments_standalone(
original_body: &[u8],
message: &mail_parser::Message<'_>,
) -> (Vec<u8>, DetachOutput) {
let mut stripped_eml = original_body.to_vec();
let mut infos = Vec::new();
let mut blobs = Vec::new();
let mut ranges: Vec<_> = message
.attachments()
.map(|att| {
(
att.raw_body_offset() as usize,
att.raw_end_offset() as usize,
att,
)
})
.collect();
ranges.sort_by(|a, b| b.0.cmp(&a.0));
for (raw_start, raw_end, att) in ranges {
let content_hash = compute_content_hash(att.contents());
let body_len = original_body.len();
let raw_start = raw_start.min(body_len);
let raw_end = raw_end.min(body_len);
let range_valid = raw_start < raw_end;
if range_valid {
blobs.push((
content_hash.clone(),
Bytes::copy_from_slice(&original_body[raw_start..raw_end]),
));
}
if range_valid {
let placeholder = format!("<<BICHON_DETACH_HASH:{}>>", &content_hash);
stripped_eml.splice(raw_start..raw_end, placeholder.as_bytes().iter().cloned());
}
infos.push(AttachmentInfo {
filename: att.attachment_name().map(|n| n.to_string()),
size: att.contents().len(),
inline: att
.content_disposition()
.map(|d| d.is_inline())
.unwrap_or_else(|| att.content_id().is_some()),
file_type: att
.content_type()
.map(|ct| {
format!(
"{}/{}",
ct.c_type.as_ref(),
ct.c_subtype.as_deref().unwrap_or("")
)
})
.unwrap_or_else(|| "application/octet-stream".to_string()),
content_id: att.content_id().map(|id| id.to_string()),
content_hash,
is_message: att.is_message(),
extracted_text: None,
extracted_page_count: None,
extracted_is_ocr: false,
});
}
(stripped_eml, DetachOutput { infos, blobs })
}
pub struct NewIndexWriter {
pub envelope_writer: Option<IndexWriter>,
pub attachment_writer: Option<IndexWriter>,
pub email_ks: Keyspace,
pub attachment_ks: Keyspace,
pending: usize,
email_buf: Vec<(String, Vec<u8>)>,
attachment_buf: Vec<(String, Vec<u8>)>,
}
//const COMMIT_THRESHOLD: usize = 500;
impl NewIndexWriter {
pub fn open(dirs: NewDirs) -> BichonResult<Self> {
// ── envelope index ──────────────────────────────────────────────
std::fs::create_dir_all(&dirs.envelope_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let envelope_index = if dirs
.envelope_dir
.read_dir()
.map(|mut d| d.next().is_none())
.unwrap_or(true)
{
Index::create_in_dir(&dirs.envelope_dir, SchemaTools::email_schema())
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?
} else {
Index::open_in_dir(&dirs.envelope_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?
};
envelope_index
.tokenizers()
.register("euro", EuroTokenizer::new());
let envelope_writer = envelope_index
.writer_with_num_threads(3, 256 * 1024 * 1024)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
// let mut merge_policy = LogMergePolicy::default();
// merge_policy.set_min_num_segments(25);
// merge_policy.set_min_layer_size(10_000);
// merge_policy.set_max_docs_before_merge(100_000);
envelope_writer.set_merge_policy(Box::new(NoMergePolicy));
// ── attachment index ─────────────────────────────────────────────
std::fs::create_dir_all(&dirs.attachment_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let attachment_index = if dirs
.attachment_dir
.read_dir()
.map(|mut d| d.next().is_none())
.unwrap_or(true)
{
Index::create_in_dir(&dirs.attachment_dir, SchemaTools::attachment_schema())
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?
} else {
Index::open_in_dir(&dirs.attachment_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?
};
attachment_index
.tokenizers()
.register("euro", EuroTokenizer::new());
let attachment_writer = attachment_index
.writer_with_num_threads(3, 256 * 1024 * 1024)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
// let mut merge_policy = LogMergePolicy::default();
// merge_policy.set_min_num_segments(25);
// merge_policy.set_min_layer_size(10_000);
// merge_policy.set_max_docs_before_merge(100_000);
attachment_writer.set_merge_policy(Box::new(NoMergePolicy));
// ── blob store ───────────────────────────────────────────────────
std::fs::create_dir_all(&dirs.storage_dir)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let db = Database::builder(&dirs.storage_dir)
.cache_size(8 * 1024 * 1024)
.journal_compression(CompressionType::None)
.max_journaling_size(64 * 1024 * 1024)
.open()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let email_ks = db
.keyspace("email", || {
KeyspaceCreateOptions::default()
.max_memtable_size(4 * 1024 * 1024)
.data_block_size_policy(BlockSizePolicy::all(4 * 1024))
.data_block_compression_policy(CompressionPolicy::all(CompressionType::Lz4))
.with_kv_separation(Some(
KvSeparationOptions::default()
.separation_threshold(1024)
.compression(CompressionType::Lz4)
.file_target_size(512 * 1024 * 1024),
))
})
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
let attachment_ks = db
.keyspace("attachments", || {
KeyspaceCreateOptions::default()
.max_memtable_size(4 * 1024 * 1024)
.data_block_size_policy(BlockSizePolicy::all(4 * 1024))
.data_block_compression_policy(CompressionPolicy::all(CompressionType::Lz4))
.with_kv_separation(Some(
KvSeparationOptions::default()
.separation_threshold(1024)
.compression(CompressionType::Lz4)
.file_target_size(512 * 1024 * 1024),
))
})
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
Ok(Self {
envelope_writer: Some(envelope_writer),
attachment_writer: Some(attachment_writer),
email_ks,
attachment_ks,
pending: 0,
email_buf: Vec::new(),
attachment_buf: Vec::new(),
})
}
pub fn ingest(
&mut self,
eml_bytes: &[u8],
account_id: u64,
mailbox_id: u64,
uid: u32,
internal_date: i64,
) -> BichonResult<()> {
let email_content_hash = compute_content_hash(eml_bytes);
let message = MessageParser::new()
.parse(eml_bytes)
.ok_or_else(|| raise_error!("failed to parse eml".into(), ErrorCode::InternalError))?;
if message.parts.is_empty() {
return Err(raise_error!(
"Malformed or completely empty EML (no parts found)".into(),
ErrorCode::InternalError
));
}
// ── text / preview ────────────────────────────────────────────────
let text = message
.body_text(0)
.map(|c| c.into_owned())
.or_else(|| {
message
.body_html(0)
.map(|html| bichon_core::utils::html::extract_text(html.into_owned()))
})
.unwrap_or_default();
let text = text.split_whitespace().collect::<Vec<_>>().join(" ");
let preview = if text.chars().count() > 100 {
text.chars().take(100).collect::<String>() + "..."
} else {
text.clone()
};
// ── headers ───────────────────────────────────────────────────────
let message_id = message
.message_id()
.map(String::from)
.unwrap_or_else(generate_message_id);
let in_reply_to = message.in_reply_to().as_text().map(String::from);
let references = extract_references(&message);
let thread_id = compute_thread_id(in_reply_to, references, &message_id);
let subject = message.subject().map(String::from).unwrap_or_default();
let date = message.date().map(|d| d.to_timestamp() * 1000).unwrap_or(0);
let internal_date = if internal_date == 0 {
date
} else {
internal_date
};
let parse_addrs = |addrs: Option<&mail_parser::Address<'_>>| {
addrs
.map(|addr| {
AddrVec::from(addr)
.0
.into_iter()
.filter_map(|a| a.address)
.collect::<Vec<_>>()
})
.unwrap_or_default()
};
let from = message
.from()
.and_then(|addr| AddrVec::from(addr).0.into_iter().next())
.and_then(|a| a.address)
.unwrap_or_else(|| "unknown".to_string());
let to = parse_addrs(message.to());
let cc = parse_addrs(message.cc());
let bcc = parse_addrs(message.bcc());
// ── detach attachments → blob ──────────────────────────────────────
let (stripped_eml, attachment_output) = detach_attachments_standalone(eml_bytes, &message);
// Buffer for bulk ingestion — sorted + flushed later.
self.email_buf
.push((email_content_hash.clone(), stripped_eml));
for (hash, data) in &attachment_output.blobs {
self.attachment_buf.push((hash.clone(), data.to_vec()));
}
// ── build envelope doc ────────────────────────────────────────────
let envelope_id = Uuid::new_v4().to_string();
let now = utc_now!();
let attachment_docs: Vec<TantivyDocument> = attachment_output
.infos
.iter()
.filter(|a| !a.inline || a.content_id.is_none())
.map(|a| {
AttachmentModel {
id: Uuid::new_v4().to_string(),
envelope_id: envelope_id.clone(),
account_id,
account_email: None,
mailbox_id,
mailbox_name: None,
subject: subject.clone(),
content_hash: a.content_hash.clone(),
from: from.clone(),
date,
ingest_at: now,
size: a.size as u64,
ext: a.get_extension(),
category: a.get_category().to_string(),
content_type: a.file_type.clone(),
shard_id: 0,
text: None,
has_text: false,
is_ocr: false,
page_count: None,
is_indexed: false,
is_message: a.is_message,
name: a.filename.clone(),
tags: None,
auto_tags: None,
}
.into_document()
})
.collect();
let envelope = Envelope {
id: envelope_id,
message_id,
account_id,
mailbox_id,
uid,
subject,
preview,
from,
to,
cc,
bcc,
date,
internal_date,
ingest_at: now,
size: eml_bytes.len() as u32,
thread_id,
attachment_count: message.attachment_count(),
regular_attachment_count: attachment_docs.len(),
tags: None,
account_email: None,
account_name: None,
mailbox_name: None,
content_hash: email_content_hash,
};
let ea = EnvelopeWithAttachments {
envelope,
attachments: Some(attachment_output.infos),
};
let envelope_doc = ea.to_document(&text, 0)?;
self.envelope_writer
.as_mut()
.unwrap()
.add_document(envelope_doc)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
for doc in attachment_docs {
self.attachment_writer
.as_mut()
.unwrap()
.add_document(doc)
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
}
self.pending += 1;
// if self.pending >= COMMIT_THRESHOLD {
// self.commit()?;
// }
Ok(())
}
/// Commit pending Tantivy documents (mid-stream) — frees the in-memory
/// term dictionary / postings that accumulate in the IndexWriter.
fn commit_tantivy(&mut self) -> BichonResult<()> {
if self.pending == 0 {
return Ok(());
}
println!("Tantivy committing... this may take 2-3 minutes, please wait.");
let start = Instant::now();
if let Some(writer) = self.envelope_writer.as_mut() {
writer
.commit()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
}
if let Some(writer) = self.attachment_writer.as_mut() {
writer
.commit()
.map_err(|e| raise_error!(format!("{e:#?}"), ErrorCode::InternalError))?;
}
println!("tantivy commit elapsed: {:#?}", start.elapsed());
tracing::info!(count = self.pending, "committed tantivy batch");
self.pending = 0;
Ok(())
}
/// Final commit + segment merge for Tantivy writers (called once at end).
pub fn finish_writers(&mut self) -> BichonResult<()> {
self.commit_tantivy()?;
for (name, writer_opt) in [
("envelope", &mut self.envelope_writer),
("attachment", &mut self.attachment_writer),
] {
if let Some(writer) = writer_opt.as_mut() {
let seg_ids = writer
.index()
.searchable_segment_ids()
.map_err(|e| raise_error!(format!("{:#?}", e), ErrorCode::InternalError))?;
println!("merging {} {} segments...", seg_ids.len(), name);
if seg_ids.len() > 1 {
let _ = writer.merge(&seg_ids);
}
}
if let Some(writer) = writer_opt.take() {
println!("waiting for {} merge to finish...", name);
let start = std::time::Instant::now();
let _ = writer.wait_merging_threads();
println!("{} merge done: {:#?}", name, start.elapsed());
}
}
Ok(())
}
/// Sort buffered (hash, data) pairs, dedup, and write via Fjall's
/// ingestion API — writes SSTables directly, bypassing memtable and WAL.
/// Also commits the Tantivy writers to bound their in-memory state.
pub fn flush_fjall_buffers(&mut self) -> BichonResult<()> {
self.commit_tantivy()?;
if !self.email_buf.is_empty() {
self.email_buf.sort_by(|a, b| a.0.cmp(&b.0));
self.email_buf.dedup_by(|a, b| a.0 == b.0);
let mut ingestion = self.email_ks.start_ingestion().map_err(|e| {
raise_error!(
format!("email ingestion start: {e:#?}"),
ErrorCode::InternalError
)
})?;
for (hash, data) in &self.email_buf {
ingestion
.write(hash.as_bytes(), data.as_slice())
.map_err(|e| {
raise_error!(
format!("email ingestion write: {e:#?}"),
ErrorCode::InternalError
)
})?;
}
ingestion.finish().map_err(|e| {
raise_error!(
format!("email ingestion finish: {e:#?}"),
ErrorCode::InternalError
)
})?;
self.email_buf.clear();
}
if !self.attachment_buf.is_empty() {
self.attachment_buf.sort_by(|a, b| a.0.cmp(&b.0));
self.attachment_buf.dedup_by(|a, b| a.0 == b.0);
let mut ingestion = self.attachment_ks.start_ingestion().map_err(|e| {
raise_error!(
format!("attachment ingestion start: {e:#?}"),
ErrorCode::InternalError
)
})?;
for (hash, data) in &self.attachment_buf {
ingestion
.write(hash.as_bytes(), data.as_slice())
.map_err(|e| {
raise_error!(
format!("attachment ingestion write: {e:#?}"),
ErrorCode::InternalError
)
})?;
}
ingestion.finish().map_err(|e| {
raise_error!(
format!("attachment ingestion finish: {e:#?}"),
ErrorCode::InternalError
)
})?;
self.attachment_buf.clear();
}
Ok(())
}
}