why tx.db.grow() is not needed in tx.root.spill()

[uvletter]

I'm learning the source code of boltdb recently. In Commit tx.root.spill() would shift tx.meta.pgid possiblely when there is no enough room in freelist, but without a tx.db.grow(). If the following freelist serialization find a feat page in the freelist, the tx.db.grow() would be eliminated, is it the expected behave?
Thanks a lot

[ptabor]

It's interesting question...

What I wanted to say is that 'grow' is postponed till tx.commitFreelist() is called:

bbolt/tx.go

Lines 239 to 240 in b654ce9

	if tx.meta.pgid > opgid {
	if err := tx.db.grow(int(tx.meta.pgid+1) * tx.db.pageSize); err != nil {

But... the commitFreelist is only called when we are NOT operating in NoFreelistSync (that's getting more popular):

bbolt/tx.go

Lines 174 to 175 in b654ce9

	if !tx.db.NoFreelistSync {
	err := tx.commitFreelist()

Intuitively the growing part should be migrated out of the 'commitFreelist' call.

I think it justifies an attempt to create a repro test for this issue.

[uvletter]

From the git history grow is introduced in boltdb/bolt#284, to solve the issue fdatasync is unsafe in some particular FS and OS version. As far as I can tell this issue has been fixed in the newer Linux kernel. As a side effect Truncate improve the performance by mitigatethe possibility of sync the metadata when flush the dirty page. But all the above will have no effect if NoFreelistSync is on...

[ahrtr]

My understanding is both spill and commitFreelist are operating on virtual memory, it's only necessary to resize the file before syncing data to db file.