Re: [RFC] Thing 1: Shardmap fox Ext4

From: Theodore Y. Ts'o
Date: Sat Nov 30 2019 - 14:03:09 EST


On Wed, Nov 27, 2019 at 08:27:59PM -0800, Daniel Phillips wrote:
> You are right that Shardmap also must update the shard fifo tail block,
> however there is only one index shard up to 64K entries, so all the new
> index entries go into the same tail block(s).

So how big is an index shard? If it is 64k entries, and each entry is
16 bytes (8 bytes hash, 8 bytes block number), then a shard is a
megabyte, right? Are entries in an index shard stored in sorted or
unsorted manner? If they are stored in an unsorted manner, then when
trying to do a lookup, you need to search all of the index shard ---
which means for a directory that is being frequently accessed, the
entire index shard has to be kept in memory, no? (Or paged in as
necessary, if you are using mmap in userspace).

> Important example: how is atomic directory commit going to work for
> Ext4?

The same way all metadata updates work in ext4. Which is to say, you
need to declare the maximum number of 4k metadata blocks that an
operation might need to change when calling ext4_journal_start() to
create a handle; and before modifying a 4k block, you need to call
ext4_journal_get_write_access(), passing in the handle and the block's
buffer_head. After modifying the block, you must call
ext4_handle_dirty_metadata() on the buffer_head. And when you are
doing with the changes in an atomic metadata operation, you call
ext4_journal_stop() on the handle.

This hasn't changed since the days of ext3 and htree.

- Ted