You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
mongo-rust-driver/src/cursor.rs

380 lines
12 KiB
Rust

//! Access to a MongoDB query cursor.
use std::iter::Iterator;
use std::ptr;
use std::thread;
use std::time::Duration;
use std::collections::VecDeque;
use mongoc::bindings;
use bson::{self,Bson,Document,oid};
use super::BsoncError;
use super::bsonc;
use super::client::Client;
use super::database::Database;
use super::flags::QueryFlag;
use super::collection::{Collection,TailOptions};
use super::CommandAndFindOptions;
use super::MongoError::ValueAccessError;
use super::Result;
#[doc(hidden)]
pub enum CreatedBy<'a> {
Client(&'a Client<'a>),
Database(&'a Database<'a>),
Collection(&'a Collection<'a>)
}
/// Provides access to a MongoDB cursor for a normal operation.
///
/// It wraps up the wire protocol negotiation required to initiate a query and
/// retrieve an unknown number of documents. Cursors are lazy, meaning that no network
/// traffic occurs until the first call to `next`. At this point various functions to get
/// information about the state of the cursor are available.
///
/// `Cursor` implements the `Iterator` trait, so you can use with all normal Rust means
/// of iteration and looping.
pub struct Cursor<'a> {
_created_by: CreatedBy<'a>,
inner: *mut bindings::mongoc_cursor_t,
tailing: bool,
tail_wait_duration: Duration,
// Become owner of bsonc because the cursor needs it
// to be allocated for it's entire lifetime
_fields: Option<bsonc::Bsonc>
}
impl<'a> Cursor<'a> {
#[doc(hidden)]
pub fn new(
created_by: CreatedBy<'a>,
inner: *mut bindings::mongoc_cursor_t,
fields: Option<bsonc::Bsonc>
) -> Cursor<'a> {
assert!(!inner.is_null());
Cursor {
_created_by: created_by,
inner: inner,
tailing: false,
tail_wait_duration: Duration::from_millis(0),
_fields: fields
}
}
fn is_alive(&self) -> bool {
assert!(!self.inner.is_null());
unsafe {
bindings::mongoc_cursor_is_alive(self.inner) == 1
}
}
fn more(&self) -> bool {
assert!(!self.inner.is_null());
unsafe {
bindings::mongoc_cursor_more(self.inner) == 1
}
}
fn error(&self) -> BsoncError {
assert!(!self.inner.is_null());
let mut error = BsoncError::empty();
unsafe {
bindings::mongoc_cursor_error(
self.inner,
error.mut_inner()
)
};
error
}
}
impl<'a> Iterator for Cursor<'a> {
type Item = Result<Document>;
fn next(&mut self) -> Option<Self::Item> {
assert!(!self.inner.is_null());
loop {
if !self.more() {
return None
}
// The C driver writes the document to memory and sets an
// already existing pointer to it.
let mut bson_ptr: *const bindings::bson_t = ptr::null();
let success = unsafe {
bindings::mongoc_cursor_next(
self.inner,
&mut bson_ptr
)
};
// Fetch error that might have occurred while getting
// the next item.
let error = self.error();
if success == 0 {
if error.is_empty() {
if self.tailing && self.is_alive() {
// Since there was no error, this is a tailing cursor
// and the cursor is alive we'll wait before trying again.
thread::sleep(self.tail_wait_duration);
continue;
} else {
// No result, no error and cursor not tailing so we must
// be at the end.
return None
}
} else {
// There was an error
return Some(Err(error.into()))
}
}
assert!(!bson_ptr.is_null());
// Parse and return bson document.
let bsonc = bsonc::Bsonc::from_ptr(bson_ptr);
match bsonc.as_document() {
Ok(document) => return Some(Ok(document)),
Err(error) => return Some(Err(error.into()))
}
}
}
}
impl<'a> Drop for Cursor<'a> {
fn drop(&mut self) {
assert!(!self.inner.is_null());
unsafe {
bindings::mongoc_cursor_destroy(self.inner);
}
}
}
/// Cursor that will reconnect and resume tailing a collection
/// at the right point if the connection fails.
///
/// This cursor will wait for new results when there are none, so calling `next`
/// is a blocking operation. If an error occurs the iterator will retry, if errors
/// keep occuring it will eventually return an error result.
pub struct TailingCursor<'a> {
collection: &'a Collection<'a>,
query: Document,
find_options: CommandAndFindOptions,
tail_options: TailOptions,
cursor: Option<Cursor<'a>>,
last_seen_id: Option<oid::ObjectId>,
retry_count: u32
}
impl<'a> TailingCursor<'a> {
#[doc(hidden)]
pub fn new(
collection: &'a Collection<'a>,
query: Document,
find_options: CommandAndFindOptions,
tail_options: TailOptions
) -> TailingCursor<'a> {
// Add flags to make query tailable
let mut find_options = find_options;
find_options.query_flags.add(QueryFlag::TailableCursor);
find_options.query_flags.add(QueryFlag::AwaitData);
TailingCursor {
collection: collection,
query: query,
find_options: find_options,
tail_options: tail_options,
cursor: None,
last_seen_id: None,
retry_count: 0
}
}
}
impl<'a> Iterator for TailingCursor<'a> {
type Item = Result<Document>;
fn next(&mut self) -> Option<Self::Item> {
loop {
// Start a scope so we're free to set the cursor to None at the end.
{
if self.cursor.is_none() {
// Add the last seen id to the query if it's present.
match self.last_seen_id.take() {
Some(id) => {
self.query.insert_bson("_id".to_string(), Bson::Document(doc!{ "$gt" => id }));
},
None => ()
};
// Set the cursor
self.cursor = match self.collection.find(&self.query, Some(&self.find_options)) {
Ok(mut c) => {
c.tailing = true;
c.tail_wait_duration = self.tail_options.wait_duration;
Some(c)
},
Err(e) => return Some(Err(e.into()))
};
}
let cursor = match self.cursor {
Some(ref mut c) => c,
None => panic!("It should be impossible to not have a cursor here")
};
match cursor.next() {
Some(next_result) => {
match next_result {
Ok(next) => {
// This was successfull, so reset retry count and return result.
self.retry_count = 0;
return Some(Ok(next))
},
Err(e) => {
// Retry if we haven't exceeded the maximum number of retries.
if self.retry_count >= self.tail_options.max_retries {
return Some(Err(e.into()))
}
}
}
},
None => ()
};
}
// We made it to the end, so we weren't able to get the next item from
// the cursor. We need to reconnect in the next iteration of the loop.
self.retry_count += 1;
self.cursor = None;
}
}
}
type DocArray = VecDeque<Document>;
type CursorId = i64;
/// BatchCursor let's you iterate though batches of results
/// in a natural way without having to deal parsing each block
/// of 100 results from mongo.
///
/// Specifically, this cursor hides the complexity of having to call
/// https://docs.mongodb.com/manual/reference/command/getMore/. This
/// allows you to have much cleaner user code. Only commands which
/// return batches work with this cursor. For example, find, aggregate,
/// and listIndexes all return batches.
pub struct BatchCursor<'a> {
cursor: Cursor<'a>,
db: &'a Database<'a>,
coll_name: String,
cursor_id: Option<CursorId>,
documents: Option<DocArray>
}
impl<'a> BatchCursor<'a> {
pub fn new(
cursor: Cursor<'a>,
db: &'a Database<'a>,
coll_name: String
) -> BatchCursor<'a> {
BatchCursor {
cursor,
db,
coll_name,
cursor_id: None,
documents: None
}
}
// internal function to reach the next batch of results from the mongo cursor
// and store them in the DocArray buffer
fn get_cursor_next(&mut self) -> Option<Result<Document>> {
let item_opt = self.cursor.next();
if let Some(item_res) = item_opt {
if let Ok(item) = item_res {
let docs_ret = batch_to_array(item);
if let Ok(docs) = docs_ret {
self.documents = docs.0;
if docs.1.is_some() {self.cursor_id = docs.1}
let res = self.get_next_doc();
if res.is_some() { return res; }
} else {
return Some(Err(docs_ret.err().unwrap()));
}
}
}
None
}
// internal function for pulling the next document from the documents buffer.
// this is the in memory representation of the documents we receive from each batch (DocArray)
fn get_next_doc(&mut self) -> Option<Result<Document>> {
if let Some(ref mut docs) = self.documents {
if docs.len() > 0 {
let doc = docs.pop_front().unwrap();
return Some(Ok(doc));
}
}
None
}
}
#[derive(Deserialize, Debug)]
#[serde(rename_all = "camelCase")]
struct CommandSimpleBatch {
id: CursorId,
first_batch: Option<DocArray>,
next_batch: Option<DocArray>
}
#[derive(Deserialize, Debug)]
struct CommandSimpleResult {
cursor: CommandSimpleBatch
}
fn batch_to_array(doc: Document) -> Result<(Option<DocArray>,Option<CursorId>)> {
let doc_result: Result<CommandSimpleResult> =
bson::from_bson(Bson::Document(doc.clone()))
.map_err(|err| {
error!("cannot read batch from db: {}", err);
ValueAccessError(bson::ValueAccessError::NotPresent)
});
doc_result.map(|v| {
if v.cursor.first_batch.is_some() {return (v.cursor.first_batch, Some(v.cursor.id));}
if v.cursor.next_batch.is_some() {return (v.cursor.next_batch, Some(v.cursor.id));}
(None,None)
})
}
impl<'a> Iterator for BatchCursor<'a> {
type Item = Result<Document>;
fn next(&mut self) -> Option<Self::Item> {
// (1) try the local document buffer
let res = self.get_next_doc();
if res.is_some() {return res;}
// (2) try next()
let res = self.get_cursor_next();
if res.is_some() {return res;}
// (3) try getMore
if let Some(cid) = self.cursor_id {
let command = doc! {
"getMore": cid as i64,
"collection": self.coll_name.clone()
};
let cur_result = self.db.command(command, None);
if let Ok(cur) = cur_result {
self.cursor = cur;
let res = self.get_cursor_next();
if res.is_some() { return res; }
}
}
None
}
}