如何通过bluestore读写分析实现长尾词优化？

本文共计245个文字，预计阅读时间需要1分钟。

plaintext写入BlueStore队列事务：通过添加事务、写入数据，根据事务大小选择小批量或大批量写入。

write

BlueStore::queue_transactions => BlueStore::_txc_add_transaction => BlueStore::_write => BlueStore::_do_write_data

=>

if (head) _do_write_small(txc, c, o, head_offset, head_length, p, wctx);

_do_write_big(txc, c, o, middle_offset, middle_length, p, wctx);

if (tail) _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx);

_txc_add_transaction 分析

1、查找CollectionRef

CollectionRef &c = cvec[op->cid];

2、循环列表，获取op

for (int pos = 0; i.have_op(); ++pos) {

Transaction::Op *op = i.decode_op();

...

}

3、获取onnode,不存在就创建一个

OnodeRef &o = ovec[op->oid];

if (!o) {

ghobject_t oid = i.get_oid(op->oid);

o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE);

}

4、分配WriteContext

int BlueStore::_do_write {

WriteContext wctx;

...

}

5、

BlueStore::_do_write_small