mirror of
https://github.com/meilisearch/MeiliSearch
synced 2025-07-03 20:07:09 +02:00
Don't iterate all indexes manually
This commit is contained in:
parent
5822764be9
commit
3db613ff77
6 changed files with 80 additions and 27 deletions
|
@ -788,15 +788,15 @@ impl IndexScheduler {
|
|||
dump_tasks.flush()?;
|
||||
|
||||
// 3. Dump the indexes
|
||||
for (uid, index) in self.index_mapper.indexes(&rtxn)? {
|
||||
self.index_mapper.try_for_each_index(&rtxn, |uid, index| -> Result<()> {
|
||||
let rtxn = index.read_txn()?;
|
||||
let metadata = IndexMetadata {
|
||||
uid: uid.clone(),
|
||||
uid: uid.to_owned(),
|
||||
primary_key: index.primary_key(&rtxn)?.map(String::from),
|
||||
created_at: index.created_at(&rtxn)?,
|
||||
updated_at: index.updated_at(&rtxn)?,
|
||||
};
|
||||
let mut index_dumper = dump.create_index(&uid, &metadata)?;
|
||||
let mut index_dumper = dump.create_index(uid, &metadata)?;
|
||||
|
||||
let fields_ids_map = index.fields_ids_map(&rtxn)?;
|
||||
let all_fields: Vec<_> = fields_ids_map.iter().map(|(id, _)| id).collect();
|
||||
|
@ -809,9 +809,10 @@ impl IndexScheduler {
|
|||
}
|
||||
|
||||
// 3.2. Dump the settings
|
||||
let settings = meilisearch_types::settings::settings(&index, &rtxn)?;
|
||||
let settings = meilisearch_types::settings::settings(index, &rtxn)?;
|
||||
index_dumper.settings(&settings)?;
|
||||
}
|
||||
Ok(())
|
||||
})?;
|
||||
|
||||
let dump_uid = started_at.format(format_description!(
|
||||
"[year repr:full][month repr:numerical][day padding:zero]-[hour padding:zero][minute padding:zero][second padding:zero][subsecond digits:3]"
|
||||
|
|
|
@ -682,18 +682,38 @@ impl IndexMapper {
|
|||
Ok(index)
|
||||
}
|
||||
|
||||
/// Return all indexes, may open them if they weren't already opened.
|
||||
pub fn indexes(&self, rtxn: &RoTxn) -> Result<Vec<(String, Index)>> {
|
||||
/// Attempts `f` for each index that exists in the index mapper.
|
||||
///
|
||||
/// It is preferable to use this function rather than a loop that opens all indexes, as a way to avoid having all indexes opened,
|
||||
/// which is unsupported in general.
|
||||
///
|
||||
/// Since `f` is allowed to return a result, and `Index` is cloneable, it is still possible to wrongly build e.g. a vector of
|
||||
/// all the indexes, but this function makes it harder and so less likely to do accidentally.
|
||||
pub fn try_for_each_index<U, V>(
|
||||
&self,
|
||||
rtxn: &RoTxn,
|
||||
mut f: impl FnMut(&str, &Index) -> Result<U>,
|
||||
) -> Result<V>
|
||||
where
|
||||
V: FromIterator<U>,
|
||||
{
|
||||
self.index_mapping
|
||||
.iter(rtxn)?
|
||||
.map(|ret| {
|
||||
ret.map_err(Error::from).and_then(|(name, _)| {
|
||||
self.index(rtxn, name).map(|index| (name.to_string(), index))
|
||||
})
|
||||
.map(|res| {
|
||||
res.map_err(Error::from)
|
||||
.and_then(|(name, _)| self.index(rtxn, name).and_then(|index| f(name, &index)))
|
||||
})
|
||||
.collect()
|
||||
}
|
||||
|
||||
/// Return the name of all indexes without opening them.
|
||||
pub fn index_names(&self, rtxn: &RoTxn) -> Result<Vec<String>> {
|
||||
self.index_mapping
|
||||
.iter(rtxn)?
|
||||
.map(|res| res.map_err(Error::from).map(|(name, _)| name.to_string()))
|
||||
.collect()
|
||||
}
|
||||
|
||||
/// Swap two index names.
|
||||
pub fn swap(&self, wtxn: &mut RwTxn, lhs: &str, rhs: &str) -> Result<()> {
|
||||
let lhs_uuid = self
|
||||
|
|
|
@ -254,6 +254,6 @@ pub fn snapshot_canceled_by(
|
|||
snap
|
||||
}
|
||||
pub fn snapshot_index_mapper(rtxn: &RoTxn, mapper: &IndexMapper) -> String {
|
||||
let names = mapper.indexes(rtxn).unwrap().into_iter().map(|(n, _)| n).collect::<Vec<_>>();
|
||||
let names = mapper.index_names(rtxn).unwrap();
|
||||
format!("{names:?}")
|
||||
}
|
||||
|
|
|
@ -541,15 +541,42 @@ impl IndexScheduler {
|
|||
///
|
||||
/// * If the index wasn't opened before, the index will be opened.
|
||||
/// * If the index doesn't exist on disk, the `IndexNotFoundError` is thrown.
|
||||
///
|
||||
/// ### Note
|
||||
///
|
||||
/// As an `Index` requires a large swath of the virtual memory address space, correct usage of an `Index` does not
|
||||
/// keep its handle for too long.
|
||||
///
|
||||
/// Some configurations also can't reasonably open multiple indexes at once.
|
||||
/// If you need to fetch information from or perform an action on all indexes,
|
||||
/// see the `try_for_each_index` function.
|
||||
pub fn index(&self, name: &str) -> Result<Index> {
|
||||
let rtxn = self.env.read_txn()?;
|
||||
self.index_mapper.index(&rtxn, name)
|
||||
}
|
||||
|
||||
/// Return and open all the indexes.
|
||||
pub fn indexes(&self) -> Result<Vec<(String, Index)>> {
|
||||
/// Return the name of all indexes without opening them.
|
||||
pub fn index_names(self) -> Result<Vec<String>> {
|
||||
let rtxn = self.env.read_txn()?;
|
||||
self.index_mapper.indexes(&rtxn)
|
||||
self.index_mapper.index_names(&rtxn)
|
||||
}
|
||||
|
||||
/// Attempts `f` for each index that exists known to the index scheduler.
|
||||
///
|
||||
/// It is preferable to use this function rather than a loop that opens all indexes, as a way to avoid having all indexes opened,
|
||||
/// which is unsupported in general.
|
||||
///
|
||||
/// Since `f` is allowed to return a result, and `Index` is cloneable, it is still possible to wrongly build e.g. a vector of
|
||||
/// all the indexes, but this function makes it harder and so less likely to do accidentally.
|
||||
///
|
||||
/// If many indexes exist, this operation can take time to complete (in the order of seconds for a 1000 of indexes) as it needs to open
|
||||
/// all the indexes.
|
||||
pub fn try_for_each_index<U, V>(&self, f: impl FnMut(&str, &Index) -> Result<U>) -> Result<V>
|
||||
where
|
||||
V: FromIterator<U>,
|
||||
{
|
||||
let rtxn = self.env.read_txn()?;
|
||||
self.index_mapper.try_for_each_index(&rtxn, f)
|
||||
}
|
||||
|
||||
/// Return the task ids matched by the given query from the index scheduler's point of view.
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue