This is the first step towards rollback, since we still don't spill
pages with WAL, we can simply invalidate page cache in case of failure.
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#1599
- Instead of using a confusing CheckpointStatus for many different things,
introduce the following statuses:
* PagerCacheflushStatus - cacheflush can result in either:
- the WAL being written to disk and fsynced
- but also a checkpoint to the main BD file, and fsyncing the main DB file
Reflect this in the type.
* WalFsyncStatus - previously CheckpointStatus was also used for this, even
though fsyncing the WAL doesn't checkpoint.
* CheckpointStatus/CheckpointResult is now used only for actual checkpointing.
- Rename HaltState to CommitState (program.halt_state -> program.commit_state)
- Make WAL a non-optional property in Pager
* This gets rid of a lot of if let Some(...) boilerplate
* For ephemeral indexes, provide a DummyWAL implementation that does nothing.
- Rename program.halt() to program.commit_txn()
- Add some documentation comments to structs and functions
Currently we are simply unable to read any WAL frames from disk
once a fresh process w/ Limbo is opened, since we never try to read
anything from disk unless we already have it in our in-memory
frame cache.
This commit implements a crude way of reading entire WAL into memory
as a single buffer and reconstructing the frame cache.
this commit changes the btree_destroy() signature to return an Option<usize>. This more closely resembles Rust semantics instead of passing a pointer to a usize.
However, I'm unsure if I'm handling the cursor result correctly
Now when dropping a table, an ephemeral table is created as a scratch table. If a root page of some other table is moved into the page occupied by the root page of the table being dropped, that row is first written into an ephemeral table. Then on a next pass, it is deleted from the schema table and then re-inserted with the new root page.
This happens during AUTOVACUUM when deleting a root page will force the last root page to move into the slot being vacated by the root page of the table being deleted
One problem we have with PageRef, is that this Page reference can be
unloaded, this means if we read the page again instead of loading the
page onto the same reference, we will have split brain of references.
To solve this we wrap PageRef in `BTreePage` so that if a page is seen
as unloaded, we will replace BTreePage::page with the newest version of
the page.
Previously, fuzz tests increase the size of page cache indefinitely,
therefore the was no problem of reaching the capacity of a page cache.
By adding transactions to fuzz tests we allow pages to remove dirty
flags once insert is finished.
After inserting a page into the wal, we dispose of the modified page.
This is unnecessary as we can simply move new page to the newest
snapshot where this page can be read.
Dirty pages can be deleted in `cacheflush`. Furthermore, there could be
multiple live references in the stack of a cursor so let's allow them to
exist while deleting.
insert() fails if key exists (there shouldn't be two) and panics if
it's different pages, and also fails if it can't make room for the page.
Replaced the limited pop_if_not_dirty() function with make_room_for().
It tries to evict many pages as requested spare capacity. It should come
handy later by resize() and Pager. make_room_for() tries to make room or
fails if it can't evict enough entries.
For make_room_for() I also tried with an all-or-nothing approach, so if
say a query requests a lot more than possible to make room for, it
doesn't evict a bunch of pages from the cache that might be useful. But
implementing this approach got very complicated since it needs to keep
exclusive PageRefs and collecting this caused segfaults. Might be worth
trying again in the future. But beware the rabbit hole.
Updated page cache test logic for new insert rules.
Updated Pager.allocate_page() to handle failure logic but needs further
work. This is to show new cache insert handling. There are many places
to update.
Left comments on callers of pager and page cache needing to update
error handling, for now.
Add error handling and results for insert(), delete(), _delete(),
_detach(), pop_if_not_dirty(), and clear.
Now these functions fail if a page is dirty, locked, or has other
references.
insert() makes room with pop_if_not_dirty() beforehand to handle
cache full and un-evictable, else it would evict this page
silently.
_delete() returns Ok when key is not present in cache and it tries
first to detach the cache entry and clean its page *before*
removing the entry from the map.
detach() checks firstt if it's possible to evict the page and if
there are no other references to the page before taking its
contents.
test_detach_via_delete() and test_detach_via_insert() fixed by
properly checking before and after dropping the page reference.
test_page_cache_fuzz() fixed by reordering and moving reference to
the page into insert.
Other page cache tests fixed to check new function results.
All page cache tests pass.
Error handling and test fixes for Pager and BTree will be added in
a subsequent commit.
Reviewable commit by commit. CI failures are not related.
Adds support for e.g. `select first_name, sum(distinct age),
count(distinct age), avg(distinct age) from users group by 1`
Implementation details:
- Creates an ephemeral index per distinct aggregate, and jumps over the
accumulation step if a duplicate is found
Closes#1507
