The compactor recently acquired the ability to consume a v2 stream. The
v2 spec requires that all streams end with a null tombstone.
`range_tombstone_assembler`, the component the compactor uses for
converting the v2 input into its v1 output enforces this with a check on
`consume_end_of_partition()`. Normally the producer of the stream the
compactor is consuming takes care of closing the active tombstone before
the stream ends. The compactor however (or its consumer) can decide to
end the consume early, e.g. to cut the current page. When this happens
the compactor must take care of closing the tombstone itself.
Furthermore it has to keep this tombstone around to re-open it on the
next page.
This patch implements this mechanism which was left out of 134601a15e.
It also adds a unit test which reproduces the problems caused by the
missing mechanism.
The compactor now tracks the last clustering position emitted. When the
page ends, this position will be used as the position of the closing
range tombstone change. This ensures the range tombstone only covers the
actually emitted range.
Fixes: #9907
Tests: unit(dev), dtest(paging_test.py, paging_additional_test.py)
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <20220114053215.481860-1-bdenes@scylladb.com>
Instead of lengthy blurbs, switch to single-line, machine-readable
standardized (https://spdx.dev) license identifiers. The Linux kernel
switched long ago, so there is strong precedent.
Three cases are handled: AGPL-only, Apache-only, and dual licensed.
For the latter case, I chose (AGPL-3.0-or-later and Apache-2.0),
reasoning that our changes are extensive enough to apply our license.
The changes we applied mechanically with a script, except to
licenses/README.md.
Closes#9937
"
With this series the mutation compactor can now consume a v2 stream. On
the output side it still uses v1, so it can now act as an online
v2->v1 converter. This allows us to push out v2->v1 conversion to as far
as the compactor, usually the next to last component in a read pipeline,
just before the final consumer. For reads this is as far as we can go,
as the intra-node ABI and hence the result-sets built are v1. For
compaction we could go further and eliminate conversion altogether, but
this requires some further work on both the compactor and the sstable
writer and so it is left to be done later.
To summarize, this patchset enables a v2 input for the compactor and it
updates compaction and single partition reads to use it.
"
* 'mutation-compactor-consume-v2/v1' of https://github.com/denesb/scylla:
table: add make_reader_v2()
querier: convert querier_cache and {data,mutation}_querier to v2
compaction: upgrade compaction::make_interposer_consumer() to v2
mutation_reader: remove unecessary stable_flattened_mutations_consumer
compaction/compaction_strategy: convert make_interposer_consumer() to v2
mutation_writer: migrate timestamp_based_splitting_writer to v2
mutation_writer: migrate shard_based_splitting_writer to v2
mutation_writer: add v2 clone of feed_writer and bucket_writer
flat_mutation_reader_v2: add reader_consumer_v2 typedef
mutation_reader: add v2 clone of queue_reader
compact_mutation: make start_new_page() independent of mutation_fragment version
compact_mutation: add support for consuming a v2 stream
compact_mutation: extract range tombstone consumption into own method
range_tombstone_assembler: add get_range_tombstone_change()
range_tombstone_assembler: add get_current_tombstone()
Said wrapper was conceived to make unmovable `compact_mutation` because
readers wanted movable consumers. But `compact_mutation` is movable for
years now, as all its unmovable bits were moved into an
`lw_shared_ptr<>` member. So drop this unnecessary wrapper and its
unnecessary usages.
Move replica-oriented classes to the replica namespace. The main
classes moved are ::database, ::keyspace, and ::table, but a few
ancillary classes are also moved. There are certainly classes that
should be moved but aren't (like distributed_loader) but we have
to start somewhere.
References are adjusted treewide. In many cases, it is obvious that
a call site should not access the replica (but the data_dictionary
instead), but that is left for separate work.
scylla-gdb.py is adjusted to look for both the new and old names.
The database, keyspace, and table classes represent the replica-only
part of the objects after which they are named. Reading from a table
doesn't give you the full data, just the replica's view, and it is not
consistent since reconciliation is applied on the coordinator.
As a first step in acknowledging this, move the related files to
a replica/ subdirectory.
"
Like flat_mutation_reader_from_fragments, this reader is also heavily
used by tests to compose a specific workload for readers above it. So
instead of converting it, we add a v2 variant and leave the v1 variant
in place.
The v2 variant was written from scratch to have built-in support for
reading in reverse. It is built-on `mutation::consume()` to avoid
duplicating the logic of consuming the contents of the mutation. To
avoid stalls, `mutation::consume()` gets support for pausing and
resuming consuming a mutation.
Tests: unit(dev)
"
* 'flat_mutation_reader_from_mutations_v2/v2' of https://github.com/denesb/scylla:
flat_mutation_reader: convert make_flat_mutation_reader_from_mutation() v2
flat_mutation_reader: extract mutation slicing into a function
mutation: consume(): make it pausable/resumable
mutation: consume(): restructure clustering iterator initialization
test/boost/mutation_test: add rebuild test for mutation::consume()
The gc_grace_seconds is a very fragile and broken design inherited from
Cassandra. Deleted data can be resurrected if cluster wide repair is not
performed within gc_grace_seconds. This design pushes the job of making
the database consistency to the user. In practice, it is very hard to
guarantee repair is performed within gc_grace_seconds all the time. For
example, repair workload has the lowest priority in the system which can
be slowed down by the higher priority workload, so that there is no
guarantee when a repair can finish. A gc_grace_seconds value that is
used to work might not work after data volume grows in a cluster. Users
might want to avoid running repair during a specific period where
latency is the top priority for their business.
To solve this problem, an automatic mechanism to protect data
resurrection is proposed and implemented. The main idea is to remove the
tombstone only after the range that covers the tombstone is repaired.
In this patch, a new table option tombstone_gc is added. The option is
used to configure tombstone gc mode. For example:
1) GC a tombstone after gc_grace_seconds
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'timeout'} ;
This is the default mode. If no tombstone_gc option is specified by the
user. The old gc_grace_seconds based gc will be used.
2) Never GC a tombstone
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'disabled'};
3) GC a tombstone immediately
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'immediate'};
4) GC a tombstone after repair
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'repair'};
In addition to the 'mode' option, another option 'propagation_delay_in_seconds'
is added. It defines the max time a write could possibly delay before it
eventually arrives at a node.
A new gossip feature TOMBSTONE_GC_OPTIONS is added. The new tombstone_gc
option can only be used after the whole cluster supports the new
feature. A mixed cluster works with no problem.
Tests: compaction_test.py, ninja test
Fixes#3560
[avi: resolve conflicts vs data_dictionary]
In the next patches we will refactor mutation::consume(). Before doing
that add another test, which rebuilds the consumed mutation, comparing
it with the original.
Add schema parameter so that:
* Caller has better control over schema -- especially relevant for
reverse reads where it is not possible to follow the convention of
passing the query schema which is reversed compared to that of the
mutations.
* Now that we don't depend on the mutations for the schema, we can lift
the restriction on mutations not being empty: this leads to safer
code. When the mutations parameter is empty, an empty reader is
created.
Add "make_" prefix to follow convention of similar reader factory
functions.
Tests: unit(dev)
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <20211115155614.363663-1-bdenes@scylladb.com>
Currently all the code operates on the range_tombstone class.
and many of those places get the range tombstone in question
from the range_tombstone_list. Next patches will make that list
carry (and return) some new object called range_tombstone_entry,
so all the code that expects to see the former one there will
need to patched to get the range_tombstone from the _entry one.
This patch prepares the ground for that by introdusing the
range_tombstone& tombstone() { return *this; }
getter on the range_tombstone itself and patching all future
users of the _entry to call .tombstone() right now.
Next patch will remove those getters together with adding the new
range_tombstone_entry object thus automatically converting all
the patched places into using the entry in a proper way.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
If x is of type std::strong_ordering, then "x <=> 0" is equivalent to
x. These no-ops were inserted during #1449 fixes, but are now unnecessary.
They have potential for harm, since they can hide an accidental of the
type of x to an arithmetic type, so remove them.
Ref #1449.
Since compaction is layered on top of sstables, let's move all compaction code
into a new top-level directory.
This change will give me extra motivation to remove all layer violations, like
sstable calling compaction-specific code, and compaction entanglement with
other components like table and storage service.
Next steps:
- remove all layer violations
- move compaction code in sstables namespace into a new one for compaction.
- move compaction unit tests into its own file
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20210707194058.87060-1-raphaelsc@scylladb.com>
The purpose of the class in question is to start sharded storage
service to make its global instance alive. I don't know when exactly
it happened but no code that instantiates this wrapper really needs
the global storage service.
Ref: #2795
tests: unit(dev), perf_sstable(dev)
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Message-Id: <20210526170454.15795-1-xemul@scylladb.com>
This warning triggers when a range for ("for (auto x : range)") causes
non-trivial copies, prompting the developer to replace with a capture
by reference. A few minor violations in the test suite are corrected.
Closes#8699
There are some pieces left doing res <=> 0 with the
res now being a strong_ordering itself. All these can
be just dropped.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The function merge_container() accepts a trichotomic comparator returning
an int. As #1449 explains, this is dangerous as it could be mistaken for
a less comparator. Switch to std::strong_ordering, but leave a compatible
merge_container() in place as it is still needed (even after this series).
This test checks that `mutation_partition::difference()` works correctly.
One of the checks it does is: m1 + m2 == m1 + (m2 - m1).
If the two mutations are identical but have compactable data, e.g. a
shadowable tombstone shadowed by a row marker, the apply will collapse
these, causing the above equality check to fail (as m2 - m1 is null).
To prevent this, compact the two input mutations.
Fixes: #8221
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <20210310141118.212538-1-bdenes@scylladb.com>
Due to small value optimizations, the removed assertions are not true in
general. Until now, atomic_cell did not use small value optimizations, but
it will after upcoming changes.
The idea of the monotonicity checking test is: try to apply
one one random partition to another random one sequentually
failing allocations. Each time allocation fails (with the
bad_alloc exception) -- check the exception guarantee is
respected, then apply (!) the very same two partitions to
each other. At the end of the test we make sure, that an
exception may pop up at any point of application and it
will be safe.
This idea is flawed currently. When verifying the guarantee
the test moves the 2nd partition and leaves it empty for the
next loop iteration. So right on the 2nd attempt to apply
partitions it becomes a no-op, doesn't fail and no more
exceptions arise.
Fix by restoring both partitions at the end of each check.
Broken since 74db08165d.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Message-Id: <20210129153641.5449-1-xemul@scylladb.com>
"
Currently we have two parallel query paths:
* database::query() -> table::query() -> data_query()
* mutation::query()
The former is used by single partition queries, the latter by range
scans, as mutation::query() is used to convert reconcilable_result to
query::result (which means it is also used in single partition queries
if it triggers read repair). This is a rather unfortunate situation as
we have two parallel implementation of the query code, which means they
are prone to diverge, and in fact they already have -- more on that
later.
This patchset aims to remedy this situation by retiring
`mutation::query()` and migrating users to an implementation based on
the "standard" query path, in other words one using the same building
blocks as the `database::query()` path. This means using
`compact_mutation` for compacting and `query_result_builder` for result
building. These components however were created to work with
`flat_mutation_reader`, however introducing a reader into this pipeline
would mean that we'd have to make all the related APIs asynchronous,
which would cause an insane amount of churn. To avoid this, this
patchset adds an API compatible `consume()` method to `mutation`, which
can accept a `compact_mutation` instance as-is. This allows an elegant
and succinct reimplementation. So far so good.
Like mentioned above, the two implementations have diverged in time, or
have been different from the start. The difference manifest when
calculating digests, more precisely in which tombstones are included in
the digest. The retired `mutation::query()` path incorporates only
non-purgeable tombstones in the digest. The standard query path however
incorporates all tombstones, even those that can be purged. After some
scrutiny however this difference proved to be completely theoretical,
as
the code path where this would matter -- converting reconcilable result
to query result -- passes min timestamp as the query time to the
compaction, so nothing is compacted and hence the difference has no
chance to manifest.
This patch-set was motivated by the desire to provide a single solution
to #7434, instead of two, one for each path.
Tests: unit(release:v2, debug:v2, dev:v3)
"
* 'unified-query-path/v3' of https://github.com/denesb/scylla:
mutation: remove now unused query() and query_compacted()
treewide: use query_mutations() instead of mutation::query()
mutation_test: test_query_digest: ensure digest is produced consistently
mutation_query: introduce query_mutation()
mutation_query: to_data_query_result(): migrate to standard query code
mutation_query: move to_data_query_result() to mutation_partition.cc
mutation: add consume()
flat_mutation_reader: move mutation consumer concepts to separate header
mutation compactor: query compaction: ignore purgeable tombstones
test_cell_external_memory_usage uses with_allocator() to observe how some
types allocate memory. However, compiler reordering (observed with clang 11
on aarch64) can move the various thread-local CQL type object initialization
into the with_allocator() scope; so any managed object allocated as part of
this initialization also gets measured, and the test fails. The code movement
is legal, as far as I can tell.
Fix this by initializing the type object early; use an atomic_thread_fence
as an optimization barrier so the compiler doesn't eliminate the or move
the early initialization.
Closes#7951
Before we retire the mutation::query() code, expand the digest test to
check that the new code replacing it produces identical digest on all
possible equivalent mutations.
measuring_allocator is a wrapper around standard_allocator, but it exposed
the default preferred_max_contiguous_allocation, not the one from
standard_allocator. Thus managed_bytes allocated in those two allocators
had fragments of different size, and their total memory usage differed,
causing test_external_memory_usage to fail if
standard_allocator::preferred_max_contiguous_allocation was changed from the
default. Fix that.
A copy/paste error means we ignore the termination of one of the
ranges. Change the comma expression to a disjunction to avoid
the unused value warning from clang.
The code is not perfect, since if the two ranges are not the same
size we'll invoke undefined behavior, but it is no worse than before
(where we ignored the comparison completely).
Clang does not yet implement p1091r3, which allows lambdas
to capture structured bindings. To accomodate it, don't
use structured bindings for variables that are later
captured.
We want to start tracking the memory consumption of mutation fragments.
For this we need schema and permit during construction, and on each
modification, so the memory consumption can be recalculated and pass to
the permit.
In this patch we just add the new parameters and go through the insane
churn of updating all call sites. They will be used in the next patch.
Not used yet, this patch does all the churn of propagating a permit
to each impl.
In the next patch we will use it to track to track the memory
consumption of `_buffer`.
Acquire a test_env and extract an sstables_manager from that, passing it
to column_familty_test_config, in preparation for losing the default
constructor of column_familty_test_config.
The test is extended for another possible corner case:
[1, NULL, 2] vs [1, 2, NULL] should have different digests.
Also, a check for legacy behavior is added.
"
While working on another patch I was getting odd compiler errors
saying that a call to ::make_shared was ambiguous. The reason was that
seastar has both:
template <typename T, typename... A>
shared_ptr<T> make_shared(A&&... a);
template <typename T>
shared_ptr<T> make_shared(T&& a);
The second variant doesn't exist in std::make_shared.
This series drops the dependency in scylla, so that a future change
can make seastar::make_shared a bit more like std::make_shared.
"
* 'espindola/make_shared' of https://github.com/espindola/scylla:
Everywhere: Explicitly instantiate make_lw_shared
Everywhere: Add a make_shared_schema helper
Everywhere: Explicitly instantiate make_shared
cql3: Add a create_multi_column_relation helper
main: Return a shared_ptr from defer_verbose_shutdown
If somebody wants to bypass proper memory accounting they should at
the very least be forced to consider if that is indeed wise and think a
second about the limit they want to apply.
This replaces a lot of make_lw_shared(schema(...)) with
make_shared_schema(...).
This makes it easier to drop a dependency on the differences between
seastar::make_shared and std::make_shared.
Signed-off-by: Rafael Ávila de Espíndola <espindola@scylladb.com>
