较长的专题文章

0x0 backgroud Even though ARIES simplifies the recovery process and allows it to be generic for all transactional operations, recovering the database to a consistent state requires undoing all operations performed by uncommitted transactions which makes the cost of recovery proportional to the work performed by these transactions. This significantly impacts database availability since recovering a long running transaction can take several hours. This paper describes the overall design of “Constant Time Recovery” (CTR)...

paper Introduction: Memory Barriers: a Hardware View for Software Hackers 0x0 why we need memory barrier In short, because reordering memory references allows much better performance, and so memory barriers are needed to force ordering in things like synchronization primitives whose correct operation depends on ordered memory references. 0x1 Cache Structure 0x11 some cases of cache miss(not important) The cache miss means that the CPU will have to wait (or be “stalled”) for hundreds of cycles while the item is fetched from memory....

0x0 Introduction A bitmap, also known as a bit array or bitset, is a data structure that represents a fixed-size sequence of bits. That is the value of the ith bit representing the existence of the the ith object. Bare bitmap can cost much memory according to the total substantial data size, even if we have stored little infomation. Roaring bitmap provide a new method to compress the bitmap structure....

Overview This chapter explains the content of clog clog(commit log), records the commit status of each transaction. The log exists both in memory mannaged by slru buffer and disk for durability. The commit status can be the four kinds below: #define TRANSACTION_STATUS_IN_PROGRESS 0x00 #define TRANSACTION_STATUS_COMMITTED 0x01 #define TRANSACTION_STATUS_ABORTED 0x02 #define TRANSACTION_STATUS_SUB_COMMITTED 0x03 In-Disk Representation Thinking that the commit status of each transaction composites an array clog[] and clog[xid] records the status, we can easily store the array to disk by the slru....

high level view See Queries in PostgreSQL: 6. Hashing One-pass hash join Note that join in PostgreSql, we scan the right relation first, which means that the right relation is the “inner relation” and the left relation is the outer one. Two-pass hash join Since we can’t allocate as much memory as we want, instead of building a hash table of the entire table, PG split the tables to several batches where all tuples have the same hash value flag....