searchengines-web/MeiliSearch
searchengines-web/MeiliSearch
1
0
Fork 0
mirror of https://github.com/meilisearch/MeiliSearch synced 2025-07-03 03:47:02 +02:00
Code Issues Releases Wiki Activity

Fix crash when batching an index swap task containing 0 swaps

Browse source
This commit is contained in:
Loïc Lecrenier 2022-10-25 10:26:51 +02:00 • committed by Clément Renault
parent 0aca5e84b9
commit 16fac10074
No known key found for this signature in database
GPG key ID: 92ADA4E935E71FA4
4 changed files with 120 additions and 51 deletions
Download patch file Download diff file Expand all files Collapse all files

73
index-scheduler/src/batch.rs
View file

@ -37,7 +37,7 @@ use roaring::RoaringBitmap;
use time::OffsetDateTime;
use uuid::Uuid;
use crate::autobatcher::BatchKind;
use crate::autobatcher::{self, BatchKind};
use crate::utils::{self, swap_index_uid_in_task};
use crate::{Error, IndexScheduler, Query, Result, TaskId};
@ -419,41 +419,47 @@ impl IndexScheduler {
)));
}
// 5. We take the next task and try to batch all the tasks associated with this index.
if let Some(task_id) = enqueued.min() {
let task = self.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
// 5. We make a batch from the unprioritised tasks. Start by taking the next enqueued task.
let task_id = if let Some(task_id) = enqueued.min() { task_id } else { return Ok(None) };
let task = self.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
// This is safe because all the remaining task are associated with
// AT LEAST one index. We can use the right or left one it doesn't
// matter.
let index_name = task.indexes().unwrap()[0];
let index_already_exists = self.index_mapper.exists(rtxn, index_name)?;
// If the task is not associated with any index, verify that it is an index swap and
// create the batch directly. Otherwise, get the index name associated with the task
// and use the autobatcher to batch the enqueued tasks associated with it
let index_tasks = self.index_tasks(rtxn, index_name)? & enqueued;
let index_name = if let Some(&index_name) = task.indexes().first() {
index_name
} else {
assert!(matches!(&task.kind, KindWithContent::IndexSwap { swaps } if swaps.is_empty()));
return Ok(Some(Batch::IndexSwap { task }));
};
// If autobatching is disabled we only take one task at a time.
let tasks_limit = if self.autobatching_enabled { usize::MAX } else { 1 };
let index_already_exists = self.index_mapper.exists(rtxn, index_name)?;
let enqueued = index_tasks
.into_iter()
.take(tasks_limit)
.map(|task_id| {
self.get_task(rtxn, task_id)
.and_then(|task| task.ok_or(Error::CorruptedTaskQueue))
.map(|task| (task.uid, task.kind))
})
.collect::<Result<Vec<_>>>()?;
let index_tasks = self.index_tasks(rtxn, index_name)? & enqueued;
if let Some((batchkind, create_index)) =
crate::autobatcher::autobatch(enqueued, index_already_exists)
{
return self.create_next_batch_index(
rtxn,
index_name.to_string(),
batchkind,
create_index,
);
}
// If autobatching is disabled we only take one task at a time.
let tasks_limit = if self.autobatching_enabled { usize::MAX } else { 1 };
let enqueued = index_tasks
.into_iter()
.take(tasks_limit)
.map(|task_id| {
self.get_task(rtxn, task_id)
.and_then(|task| task.ok_or(Error::CorruptedTaskQueue))
.map(|task| (task.uid, task.kind))
})
.collect::<Result<Vec<_>>>()?;
if let Some((batchkind, create_index)) =
autobatcher::autobatch(enqueued, index_already_exists)
{
return self.create_next_batch_index(
rtxn,
index_name.to_string(),
batchkind,
create_index,
);
}
// If we found no tasks then we were notified for something that got autobatched
@ -1042,9 +1048,8 @@ impl IndexScheduler {
for task_id in to_delete_tasks.iter() {
let task = self.get_task(wtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
if let Some(task_indexes) = task.indexes() {
affected_indexes.extend(task_indexes.into_iter().map(|x| x.to_owned()));
}
affected_indexes.extend(task.indexes().into_iter().map(|x| x.to_owned()));
affected_statuses.insert(task.status);
affected_kinds.insert(task.kind.as_kind());
// Note: don't delete the persisted task data since