mirrors/scylladb
1
0
Fork 0
mirror of https://github.com/scylladb/scylladb.git synced 2026-04-21 00:50:35 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
9fc7f8a5ed2b045fb62fa3bca8a689f504e057df
scylladb/sstables/row.cc
Nadav Har'El b7e29691c2 sstables: avoid index and data file over-reads 
When we do a streaming read that knows the expected *end* position of the
read, we can use a large read-ahead buffer, and at the same time, stop
reading at exactly the intended end (or small rounding of it to the DMA
block size) and not waste resources blindly reading a large amount of data
after the end just to fill the read-ahead buffer.

The sstable reading code, both for reading the data file and the index file,
created a file input stream without specifiying its end, thereby losing
this optimization - so when a large buffer was used, we would get a large
over-read. This patch fixes this, so sstable data file and index file are
read using a file input stream which is a ware of its end.

Fixes #964.

Note that this patch does not change the behavior when reading a
*compressed* data file. For compressed read, we did not have the problem
of over-read in the first place, because chunks are read one by one.
But we do have other sources of inefficiencies there (stemming, again,
from the fact that the compressed chunks are read one by one), and I
opened a separate issue #992 for that.

Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <1457219304-12680-1-git-send-email-nyh@scylladb.com>
2016-03-08 17:26:10 +02:00

388 lines
14 KiB
C++
Raw Blame History

/*
* Copyright 2015 Cloudius Systems
*/
/*
* This file is part of Scylla.
*
* Scylla 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.
*
* Scylla 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 General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with Scylla. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sstables.hh"
#include "consumer.hh"
namespace sstables {
// data_consume_rows_context remembers the context that an ongoing
// data_consume_rows() future is in.
class data_consume_rows_context : public data_consumer::continuous_data_consumer<data_consume_rows_context> {
private:
enum class state {
ROW_START,
ROW_KEY_BYTES,
DELETION_TIME,
DELETION_TIME_2,
DELETION_TIME_3,
ATOM_START,
ATOM_START_2,
ATOM_NAME_BYTES,
ATOM_MASK,
ATOM_MASK_2,
EXPIRING_CELL,
EXPIRING_CELL_2,
EXPIRING_CELL_3,
CELL,
CELL_2,
CELL_VALUE_BYTES,
CELL_VALUE_BYTES_2,
RANGE_TOMBSTONE,
RANGE_TOMBSTONE_2,
RANGE_TOMBSTONE_3,
RANGE_TOMBSTONE_4,
RANGE_TOMBSTONE_5,
} _state = state::ROW_START;
row_consumer& _consumer;
temporary_buffer<char> _key;
temporary_buffer<char> _val;
// state for reading a cell
bool _deleted;
uint32_t _ttl, _expiration;
static inline bytes_view to_bytes_view(temporary_buffer<char>& b) {
// The sstable code works with char, our "bytes_view" works with
// byte_t. Rather than change all the code, let's do a cast...
using byte = bytes_view::value_type;
return bytes_view(reinterpret_cast<const byte*>(b.get()), b.size());
}
public:
bool non_consuming() const {
return (((_state == state::DELETION_TIME_3)
|| (_state == state::CELL_VALUE_BYTES_2)
|| (_state == state::ATOM_START_2)
|| (_state == state::ATOM_MASK_2)
|| (_state == state::EXPIRING_CELL_3)) && (_prestate == prestate::NONE));
}
// process() feeds the given data into the state machine.
// The consumer may request at any point (e.g., after reading a whole
// row) to stop the processing, in which case we trim the buffer to
// leave only the unprocessed part. The caller must handle calling
// process() again, and/or refilling the buffer, as needed.
row_consumer::proceed process_state(temporary_buffer<char>& data) {
#if 0
// Testing hack: call process() for tiny chunks separately, to verify
// that primitive types crossing input buffer are handled correctly.
constexpr size_t tiny_chunk = 1; // try various tiny sizes
if (data.size() > tiny_chunk) {
for (unsigned i = 0; i < data.size(); i += tiny_chunk) {
auto chunk_size = std::min(tiny_chunk, data.size() - i);
auto chunk = data.share(i, chunk_size);
if (process(chunk) == row_consumer::proceed::no) {
data.trim_front(i + chunk_size - chunk.size());
return row_consumer::proceed::no;
}
}
data.trim(0);
return row_consumer::proceed::yes;
}
#endif
switch (_state) {
case state::ROW_START:
// read 2-byte key length into _u16
if (read_16(data) != read_status::ready) {
_state = state::ROW_KEY_BYTES;
break;
}
case state::ROW_KEY_BYTES:
// After previously reading 16-bit length, read key's bytes.
if (read_bytes(data, _u16, _key) != read_status::ready) {
_state = state::DELETION_TIME;
break;
}
case state::DELETION_TIME:
if (read_32(data) != read_status::ready) {
_state = state::DELETION_TIME_2;
break;
}
// fallthrough
case state::DELETION_TIME_2:
if (read_64(data) != read_status::ready) {
_state = state::DELETION_TIME_3;
break;
}
// fallthrough
case state::DELETION_TIME_3: {
deletion_time del;
del.local_deletion_time = _u32;
del.marked_for_delete_at = _u64;
_consumer.consume_row_start(to_bytes_view(_key), del);
// after calling the consume function, we can release the
// buffers we held for it.
_key.release();
_state = state::ATOM_START;
}
case state::ATOM_START:
if (read_16(data) == read_status::ready) {
if (_u16 == 0) {
// end of row marker
_state = state::ROW_START;
if (_consumer.consume_row_end() ==
row_consumer::proceed::no) {
return row_consumer::proceed::no;
}
} else {
_state = state::ATOM_NAME_BYTES;
}
} else {
_state = state::ATOM_START_2;
}
break;
case state::ATOM_START_2:
if (_u16 == 0) {
// end of row marker
_state = state::ROW_START;
if (_consumer.consume_row_end() ==
row_consumer::proceed::no) {
return row_consumer::proceed::no;
}
} else {
_state = state::ATOM_NAME_BYTES;
}
break;
case state::ATOM_NAME_BYTES:
if (read_bytes(data, _u16, _key) != read_status::ready) {
_state = state::ATOM_MASK;
break;
}
case state::ATOM_MASK:
if (read_8(data) != read_status::ready) {
_state = state::ATOM_MASK_2;
break;
}
// fallthrough
case state::ATOM_MASK_2: {
auto mask = _u8;
enum mask_type {
DELETION_MASK = 0x01,
EXPIRATION_MASK = 0x02,
COUNTER_MASK = 0x04,
COUNTER_UPDATE_MASK = 0x08,
RANGE_TOMBSTONE_MASK = 0x10,
};
if (mask & RANGE_TOMBSTONE_MASK) {
_state = state::RANGE_TOMBSTONE;
} else if (mask & COUNTER_MASK) {
// FIXME: see ColumnSerializer.java:deserializeColumnBody
throw malformed_sstable_exception("FIXME COUNTER_MASK");
} else if (mask & EXPIRATION_MASK) {
_deleted = false;
_state = state::EXPIRING_CELL;
} else {
// FIXME: see ColumnSerializer.java:deserializeColumnBody
if (mask & COUNTER_UPDATE_MASK) {
throw malformed_sstable_exception("FIXME COUNTER_UPDATE_MASK");
}
_ttl = _expiration = 0;
_deleted = mask & DELETION_MASK;
_state = state::CELL;
}
break;
}
case state::EXPIRING_CELL:
if (read_32(data) != read_status::ready) {
_state = state::EXPIRING_CELL_2;
break;
}
// fallthrough
case state::EXPIRING_CELL_2:
_ttl = _u32;
if (read_32(data) != read_status::ready) {
_state = state::EXPIRING_CELL_3;
break;
}
// fallthrough
case state::EXPIRING_CELL_3:
_expiration = _u32;
_state = state::CELL;
case state::CELL: {
if (read_64(data) != read_status::ready) {
_state = state::CELL_2;
break;
}
}
case state::CELL_2:
if (read_32(data) != read_status::ready) {
_state = state::CELL_VALUE_BYTES;
break;
}
case state::CELL_VALUE_BYTES:
if (read_bytes(data, _u32, _val) == read_status::ready) {
// If the whole string is in our buffer, great, we don't
// need to copy, and can skip the CELL_VALUE_BYTES_2 state.
//
// finally pass it to the consumer:
if (_deleted) {
if (_val.size() != 4) {
throw malformed_sstable_exception("deleted cell expects local_deletion_time value");
}
deletion_time del;
del.local_deletion_time = consume_be<uint32_t>(_val);
del.marked_for_delete_at = _u64;
_consumer.consume_deleted_cell(to_bytes_view(_key), del);
} else {
_consumer.consume_cell(to_bytes_view(_key),
to_bytes_view(_val), _u64, _ttl, _expiration);
}
// after calling the consume function, we can release the
// buffers we held for it.
_key.release();
_val.release();
_state = state::ATOM_START;
} else {
_state = state::CELL_VALUE_BYTES_2;
}
break;
case state::CELL_VALUE_BYTES_2:
if (_deleted) {
if (_val.size() != 4) {
throw malformed_sstable_exception("deleted cell expects local_deletion_time value");
}
deletion_time del;
del.local_deletion_time = consume_be<uint32_t>(_val);
del.marked_for_delete_at = _u64;
_consumer.consume_deleted_cell(to_bytes_view(_key), del);
} else {
_consumer.consume_cell(to_bytes_view(_key),
to_bytes_view(_val), _u64, _ttl, _expiration);
}
// after calling the consume function, we can release the
// buffers we held for it.
_key.release();
_val.release();
_state = state::ATOM_START;
break;
case state::RANGE_TOMBSTONE:
if (read_16(data) != read_status::ready) {
_state = state::RANGE_TOMBSTONE_2;
break;
}
case state::RANGE_TOMBSTONE_2:
// read the end column into _val.
if (read_bytes(data, _u16, _val) != read_status::ready) {
_state = state::RANGE_TOMBSTONE_3;
break;
}
case state::RANGE_TOMBSTONE_3:
if (read_32(data) != read_status::ready) {
_state = state::RANGE_TOMBSTONE_4;
break;
}
case state::RANGE_TOMBSTONE_4:
if (read_64(data) != read_status::ready) {
_state = state::RANGE_TOMBSTONE_5;
break;
}
case state::RANGE_TOMBSTONE_5:
{
deletion_time del;
del.local_deletion_time = _u32;
del.marked_for_delete_at = _u64;
_consumer.consume_range_tombstone(to_bytes_view(_key),
to_bytes_view(_val), del);
_key.release();
_val.release();
_state = state::ATOM_START;
break;
}
default:
throw malformed_sstable_exception("unknown state");
}
return row_consumer::proceed::yes;
}
data_consume_rows_context(row_consumer& consumer,
input_stream<char> && input, uint64_t maxlen) :
continuous_data_consumer(std::move(input), maxlen)
, _consumer(consumer) {
}
void verify_end_state() {
if (_state != state::ROW_START || _prestate != prestate::NONE) {
throw malformed_sstable_exception("end of input, but not end of row");
}
}
};
// data_consume_rows() and data_consume_rows_at_once() both can read just a
// single row or many rows. The difference is that data_consume_rows_at_once()
// is optimized to reading one or few rows (reading it all into memory), while
// data_consume_rows() uses a read buffer, so not all the rows need to fit
// memory in the same time (they are delivered to the consumer one by one).
class data_consume_context::impl {
private:
std::unique_ptr<data_consume_rows_context> _ctx;
public:
impl(row_consumer& consumer,
input_stream<char>&& input, uint64_t maxlen) :
_ctx(new data_consume_rows_context(consumer, std::move(input), maxlen)) { }
future<> read() {
return _ctx->consume_input(*_ctx);
}
};
data_consume_context::~data_consume_context() = default;
data_consume_context::data_consume_context(data_consume_context&& o) noexcept
: _pimpl(std::move(o._pimpl))
{ }
data_consume_context& data_consume_context::operator=(data_consume_context&& o) noexcept {
_pimpl = std::move(o._pimpl);
return *this;
}
data_consume_context::data_consume_context(std::unique_ptr<impl> p) : _pimpl(std::move(p)) { }
future<> data_consume_context::read() {
return _pimpl->read();
}
data_consume_context sstable::data_consume_rows(
row_consumer& consumer, uint64_t start, uint64_t end) {
// TODO: The second "end - start" below is redundant: The first one tells
// data_stream() to stop at the "end" byte, which allows optimal read-
// ahead and avoiding over-read at the end. The second one tells the
// consumer to stop at exactly the same place, and forces the consumer
// to maintain its own byte count.
return std::make_unique<data_consume_context::impl>(
consumer, data_stream(start, end - start, consumer.io_priority()), end - start);
}
data_consume_context sstable::data_consume_rows(row_consumer& consumer) {
return data_consume_rows(consumer, 0, data_size());
}
future<> sstable::data_consume_rows_at_once(row_consumer& consumer,
uint64_t start, uint64_t end) {
return data_read(start, end - start, consumer.io_priority()).then([&consumer]
(temporary_buffer<char> buf) {
data_consume_rows_context ctx(consumer, input_stream<char>(), -1);
ctx.process(buf);
ctx.verify_end_state();
});
}
}
Reference in New Issue View Git Blame Copy Permalink