The semantics of Scylla's materialized views may vary depending on how their
primary keys correspond to the base table's one. One of the differences is
how we handle writes to columns in the base table that are not selected by
a view:
* Case 1: The view's PK is a permutation of the base table's PK:
Since the view's primary key cannot be changed in an update, a row in
the view remains alive as long as the corresponding row in the base table
is alive.
The tricky part comes when the base table has columns that are NOT selected
by the view. CQL3 used to not allow for defining a table that didn't have
any other columns besides its primary key. Also, when inserting a row into
a table, it was mandatory to provide at least one value aside from the
primary key. At some point it changed [1] and the implementation of the
solution relied on the notion of the row marker.
Putting the details aside, consider the following scenario:
(i) the base table has a primary key consisting of columns
c_1, ..., c_k, and it has regular columns rc_1, ..., rc_n,
(ii) the primary key of an MV defined on that table consists of
a permutation of c_1, ..., c_k. The MV doesn't select at least
one of the regular columns of the base table. Without loss of
generality, let that unselected column be rc_1.
(iii) the base table has a row R whose only non-null value is the one
in the regular column rc_1.
Now, what will R correspond to in the MV? The base table doesn't have a row
marker, but all of its regular columns in the MV will be NULLs. That's NOT
allowed.
To solve that problem, all unselected columns have corresponding virtual
columns in the MV; the only information they provide is whether there is
a value in the base table or not. This way, the MV knows if a row is still
alive or not.
For that reason, we send view updates to virtual columns in the following
cases:
(i) the value in the column changes from NULL to a value, i.e. it's
created,
(ii) the value in the column exists, but its TTL has been updated.
* Case 2: The view's PK has one more column that the base table's one:
Since the primary key of the view has a regular column C from the base
table, it is guaranteed that if there's a row in the MV, the corresponding
row in the base table can remain alive: since C is part of the view's PK,
it must have a value, so the row in the base table has a value in C too.
The problem with virtual columns from the previous case doesn't manifest
in this one. The liveness of the cell in C determines the liveness of
the whole row in the view.
The semantics gets more complex, but the conclusion is this: in case 1,
virtual columns exist and we may need to generate view updates for them,
while in case 2 virtual columns do NOT exist and so we don't generate
view updates for them.
What changes in this patch is we adjust the code to it. If a view has
a regular column from the base table as part of its primary key, we
no longer emit view updates when we change a column unselected by that
view. It is purely an OPTIMIZATION change.
[1]: https://issues.apache.org/jira/browse/CASSANDRA-4361Fixesscylladb/scylladb#21652Closesscylladb/scylladb#21653
Schema of system tables is defined statically and table_schema_version needs to be explicitly set in code like this:
```
builder.with_version(system_keyspace::generate_schema_version(table_id, version_offset));
```
Whenever schema is changed, the schema version needs to change, otherwise we hit undefined behavior when trying to interpret mutation data created with the old schema using the new schema.
It's not obvious that one needs to do that and developers often forget to do that. There were several instances of mistakes of omission, some caught during review, some not, e.g.: 31ea74b96e.
This patch changes definitions to call the new `schema_builder::with_hash_version()`, which will make the schema builder compute version from schema definition so that changes of the schema will automatically change the version. This way we no longer rely on the developer to remember to bump the version offset.
All nodes should arrive at the same version, which is verified by existing `test_group0_schema_versioning` and a new unit test: `test_system_schema_version_is_stable`.
Closesscylladb/scylladb#21602
* github.com:scylladb/scylladb:
system_tables: Compute schema version automatically
schema_builder: Introduce with_hash_version()
schema: Store raw_view_info in schema::raw_schema
schema: Remove dead comment
hashing: Add hasher for unordered_map
hashing: Add hasher for unique_ptr
hashing: Add hasher for double
[avi: add missing include <memory> to hashing.hh]
This patch introduces the skip_when_empty flag to all CQL counters that
previously lacked this setting.
The skip_when_empty flag is a metric optimization that prevents
reporting on counters that have never been used. Once a counter has been
used (i.e., it holds a positive value), it will continue to be reported
consistently from that point onward.
Fixes#21046
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Closesscylladb/scylladb#21565
instead of making a copy of the warnings vector, make the warnings a non const in prepared_statement
and move the warnings vector to execute_maybe_with_guard
Closesscylladb/scylladb#20361Closesscylladb/scylladb#21083
More gossiper cleanups that accumulated since the previous one.
* 'gleb/more-gossip-cleanup-v2' of github.com:scylladb/scylla-dev:
gossiper: replace milliseconds with seconds where appropriate
gossiper: simplify failure_detector_loop loop a bit
gossiper: use fmt library to format time
gossiper: drop on_success callback from mutate_live_and_unreachable_endpoints
gossiper: remove code duplication between shadow round and regular path when state is applied
gossiper: remove remnants of old shadow round
gossiper: fix indentation after the last patch
gossiper: co-routinize do_shadow_round
Remove the `--python` option which was originally added in 780d9a26b2 to support
CentOS's non-standard python3 path (`/usr/bin/python3.4`).
Since we now:
- Build using a Fedora-based container with standard python3 path
- Use properly configured shebangs in build scripts
- Set correct executable permissions on Python scripts
This change:
1. Removes the `--python` command line option
2. Updates build rules to execute Python scripts directly instead of via interpreter
This simplifies the build system and reduces differences between CMake and
configure.py-generated rules.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#21607
As reported by @Deexie, during the process of opening backport PRs in https://github.com/scylladb/scylladb/pull/21616, No invite emails were sent, causing a lack of permissions for the backport PR branch
The check if `has_in_collaborators(pr.user.login)` was pointing to `scylladb/scylladb` instead of `scylladbbot/scylladb`, fixing it
I also moved the collaborator check to an early stage, before trying to open a backport PR
Closesscylladb/scylladb#21645
This adds a new tablet migration kind: repair. It allows tablet repair
scheduler to use this migration kind to schedule repair jobs.
The current repair scheduler implementation does the following:
- A tablet is picked to be repaired when is requested by user
- The tablet repair can be scheduled along with tablet migration and
rebuild. It runs in the tablet_migration track.
- Repair jobs are scheduled in a smart way so that at any point in time,
there are no more than configured jobs per shard, which is similar to
scylla manager's control.
New feature. No backport is needed.
Closesscylladb/scylladb#21088
* github.com:scylladb/scylladb:
test: Add tests for tablet repair scheduler
repair: Add restful API for tablet repair
repair: Add tablet repair scheduler internal API support
docs: Update system_keyspace.md for tablet repair related info
docs: Add docs for tablet repair migration
repair: Add core tablet repair scheduler support
messaging_service: Introduce TABLET_REPAIR verb
tablet_allocator: Introduce stream_weight for tablet_migration_streaming_info
network_topology_strategy: Preserve fields of task_info in reallocate_tablets
now that we are allowed to use C++23. we now have the luxury of using `std::ranges::find_if`.
in this change, we:
- replace `boost::find_if` with `std::ranges::find_if`
- remove all `#include <boost/range/algorithm/find_if.hpp>`
to reduce the dependency to boost for better maintainability, and leverage standard library features for better long-term support.
this change is part of our ongoing effort to modernize our codebase and reduce external dependencies where possible.
---
it's a cleanup, hence no need to backport.
Closesscylladb/scylladb#21495
* github.com:scylladb/scylladb:
treewide: replace boost::find_if with std::ranges::find_if
counters: replace boost::find_if with std::ranges::find_if
combine.hh: use std::iter_const_reference_t when appropriate
Currently, PER PARTITION LIMIT is not implemented for aggregates and queries can result in more rows than expected from the same partition.
Instrument the result_set_builder class so that it can enforce PER PARTITION LIMIT for aggregate queries, specifically:
- add per_partition_limit to the result_set_builder
- expose the number of input rows in the selector
result_set_builder gets two new functions handling partition start and end:
- accept_partition_end for notifying that a partition has been finished. This is also called when a page ends, so we cannot simply flush here, as a naive implementation could do.
- accept_new_partition, where we flush_selectors() if it's indeed a new partition (and not a continuation of the previous) and the query has a grouping: we don't want to flush on new partition in a query like SELECT COUNT(*) FROM foo;
Fixes#5363Closesscylladb/scylladb#21125
* github.com:scylladb/scylladb:
test: enable PER PARTIION LIMIT + GROUP BY tests
cql3: respect PER PARTITION LIMIT for aggregates
cql3: selection: count input rows in the selector
cql3: selection: pass per partition limit to the result_set_builder
cql3: show different messages for LIMIT and PER PARTITION LIMIT in get_limit
Currently, task_manager_module::abort_all_repairs marks top-level repairs as aborted (but does not abort them) and aborts all existing shard tasks.
A running repair checks whether its id isn't contained in _aborted_pending_repairs and then proceeds to create shard tasks. If abort_all_repairs is executed after _aborted_pending_repairs is checked but before shard tasks are created, then those new tasks won't be aborted. The issue is the most severe for tablet_repair_task_impl that checks the _aborted_pending_repairs content from different shards, that do not see the top-level task. Hence the repair isn't stopped but it creates shard repair tasks on all shards but the one that initialized repair.
Abort top-level tasks in abort_all_repairs. Fix the shard on which the task abort is checked.
Fixes: #21612.
Needs backport to 6.1 and 6.2 as they contain the bug.
Closesscylladb/scylladb#21616
* github.com:scylladb/scylladb:
test: add test to check if repair is properly aborted
repair: add shard param to task_manager_module::is_aborted
repair: use task abort source to abort repair
repair: drop _aborted_pending_repairs and utilize tasks abort mechanism
repair: fix task_manager_module::abort_all_repairs
Those internal APIs allow to add / del a tablet repair request and
config the tablet repair scheduler.
It can be used by task manager or plain restful api.
This adds a new tablet migration kind: repair. It allows tablet repair
scheduler to use this migration kind to schedule repair jobs.
The current repair scheduler implementation does the following:
- A tablet is picked to be repaired when the time since last repair is
bigger than a threshold (auto repair mode) or it is requested by user
(manual repair mode)
- The tablet repair can be scheduled along with tablet migration and
rebuild. It runs in the tablet_migration track.
- Repair jobs are scheduled in a smart way so that at any point in time,
there are no more than configured jobs per shard, which is similar to
scylla manager's control.
In this patch, both the manual repair and the auto repair are not
enabled yet.
Before these changes, we didn't wait for the materialized views to
finish building before writing to the base table. That led to generating
an additional view update, which, in turn, led to test failures.
The scenario corresponding to the summary above looked like this:
1. The test creates an empty table and MVs on it.
2. The view builder starts, but it doesn't finish immediately.
3. The test performs mutations to the base table. Since the views
already exist, view updates are generated.
4. Finally, the view builder finishes. It notices that the base
table has a row, so it generates a view update for it because
it doesn't notice that we already have data in the view.
We solve it by explicitly waiting for both views to finish building
and only then start writing to the base table.
Additionally, we also fix a lifetime issue of the row the test revolves
around, further stabilizing CI.
Fixes https://github.com/scylladb/scylladb/issues/20889
Backport: These changes have no semantic effect on the codebase,
but they stabilize CI, so we want to backport them to the maintained
versions of Scylla.
Closesscylladb/scylladb#21632
* github.com:scylladb/scylladb:
test/boost/view_schema_test.cc: Increase TTL in test_view_update_generating_writetime
test/boost/view_schema_test.cc: Wait for views to build in test_view_update_generating_writetime
The auxiliary function `eventually()` (defined in `test/lib/eventually.hh`)
tries to execute a passed function. If it throws, `eventually()` sleeps
for `2^#previous_attempts` milliseconds and tries to perform it again.
The default limit of attempts is 17.
In `test_view_update_generating_writetime`, right before the last test
case, we perform:
```cql
UPDATE t USING TTL 10 AND TIMESTAMP 8 SET g=40 WHERE k=1 AND c=1;
```
The test case itself executes:
```cql
SELECT WRITETIME(g) FROM t;
```
and asserts that the result of the query is equal to 8, i.e. it
corresponds to the timestamp of the last write to the table `t`.
However, if the test case keeps failing, then during its 14th attempt
(so affter sleeping for at least `2^14 - 1` milliseconds, which amounts
to about 16 seconds), we'll observe the following error:
```
[Exception] - std::runtime_error: Expected row not found: [0000000000000008] not in {result_message::rows {row: null}}
```
The reason behind it is the specified TTL is too short. 10 seconds will
have already passed before the 14th attempt, so the value in the column
`g` will be `NULL` again. In particular, the `WRITETIME(g)` will no
longer be equal to `8`.
To solve that issue, we change the TTL in the CQL statement to 300.
The time spent on 17 loops of `eventually()` amounts to about
`2^18 - 1` milliseconds, which is about 263 seconds. That's why
setting the TTL to 300 seconds should be enough to prevent the error
from occurring.
Before these changes, we didn't wait for the materialized views to
finish building before writing to the base table. That led to generating
an additional view update, which, in turn, led to test failures.
The scenario corresponding to the summary above looked like this:
1. The test creates an empty table and MVs on it.
2. The view builder starts, but it doesn't finish immediately.
3. The test performs mutations to the base table. Since the views
already exist, view updates are generated.
4. Finally, the view builder finishes. It notices that the base
table has a row, so it generates a view update for it because
it doesn't notice that we already have data in the view.
We solve it by explicitly waiting for both views to finish building
and only then start writing to the base table.
Fixesscylladb/scylladb#20889
Alternator's "/localnodes" HTTP requests is supposed to return the list
of nodes in the local DC to which the user can send requests.
Before commit bac7c33313 we used the
gossiper is_alive() method to determine if a node should be returned.
That commit changed the check to is_normal() - because a node can be
alive but in non-normal (e.g., joining) state and not ready for
requests.
However, it turns out that checking is_normal() is not enough, because
if node is stopped abruptly, other nodes will still consider it "normal",
but down (this is so-called "DN" state). So we need to check **both**
is_alive() and is_normal().
This patch also adds a test reproducing this case, where a node is
shut down abruptly. Before this patch, the test failed ("/localnodes"
continued to return the dead node), and after it it passes.
Fixes#21538
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#21540
Adding `Fixes` validation to a PR when backport labels were added. When the auto backport process triggers (after promotion), we will ensure each PR with backport/x.y label also has in the PR body a `Fixes` reference to an issue
Fixes: https://github.com/scylladb/scylladb/issues/20021Closesscylladb/scylladb#21563
Change e3e8a94c9a changed
the semantics of the enable_tablets config option,
but updating that in the option documentation in scylla.yaml
was missed.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Closesscylladb/scylladb#21614
This change was created in the same spirit of f8221b960f.
The S3ProxyServer (introduced in 8919e0abab) currently prints its
status directly to stdout, which can be distracting when reviewing test
results. For example:
```console
$ ./test.py --mode release object_store/test_backup::test_simple_backup_and_restore
Found 1 tests.
Setting minio proxy random seed to 1731924995
Starting S3 proxy server on ('127.193.179.2', 9002)
================================================================================
[N/TOTAL] SUITE MODE RESULT TEST
------------------------------------------------------------------------------
[1/1] object_store release [ PASS ] object_store.test_backup.1
Stopping S3 proxy server
------------------------------------------------------------------------------
CPU utilization: 3.1%
```
Move these messages to use proper logging to give developers more control
over their visibility:
- Make logger parameter mandatory in S3ProxyServer constructor
- Route "Stopping S3 proxy" message through the provided logger
- Add --log-level option to the standalone proxy server launcher
The message is now hidden:
```console
$ ./test.py --mode release object_store/test_backup::test_simple_backup_and_restore
Found 1 tests.
================================================================================
[N/TOTAL] SUITE MODE RESULT TEST
------------------------------------------------------------------------------
[1/1] object_store release [ PASS ] object_store.test_backup.1
------------------------------------------------------------------------------
CPU utilization: 4.1%
```
---
this change improves the developer experience, hence no need to backport.
Closesscylladb/scylladb#21610
* github.com:scylladb/scylladb:
test: route S3 Proxy server messages through logger
test: s3_proxy: remove unused method
now that we are allowed to use C++23. we now have the luxury of using
`std::ranges::find_if`.
in this change, we:
- replace `boost::find_if` with `std::ranges::find_if`
- remove all `#include <boost/range/algorithm/find_if.hpp>`
to reduce the dependency to boost for better maintainability, and
leverage standard library features for better long-term support.
this change is part of our ongoing effort to modernize our codebase
and reduce external dependencies where possible.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
std::ranges allows us to create a range from a pair of iterators.
but the iterator has to fulfill the concept of `std::semiregular`.
in order to reduce the header dependency on boost, we need to
make `basic_counter_cell_view::shard_iterator` to support
`std::semiregular`.
in this change:
- define a default constructor for
`basic_counter_cell_view::shard_iterator`, so that the iterator
satisfies the constraints of `std::semiregular`, as required by
C++20's forward_iterator concept. please note, despite that
the standard requires the iterator to be `std::semiregular`, but
the iterator created by default constructor is not evaluated in
production. sometimes, the standard algorithms just need to
store/create itermediate iterators or to represent a "singular"
state for iterator. a use case is an empty container.
- change `basic_counter_cell_view::shard_iterator::reference` so
its dereference returns a rvalue instead of a reference. because
per C++20 standard, the dereference of a forward_iterator should
be stable, but we were returning a reference / pointer referencing
a member variable of the iterator. so once the iterator is destructed,
the returned reference / pointer would be invalidated. so we have to
return a value to fulfill the requiremend of forward_iterator. this
change also fulfills the requirement of `same_as<iter_reference_t<It>,
iter_reference_t<const It>>`, which a part of the
`indirectly_readable` requirement.
- let `basic_counter_cell_view::shards()` return a subrange
- let `basic_counter_shard_view::swap_value_and_clock()` accepts
a plain value instead of a reference. because the dereference of
the iterator does not return a reference anymore. and the returned
type is a lightweighted "view", so the performance penality is
negligible.
- use ranges libraries when appropriate in this header.
this change is part of our ongoing effort to modernize our codebase
and reduce external dependencies where possible.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
before this change, we assumed that the dereference types of
the given `InputIterator1` and `InputIterator2` are always
references. but this does not hold if the `operator*` returns
a rvalue, as in the C++20 standard, unlike the LegacyForwardIterator
requirement, `std::forward_iterator` does not requires
dereference to return a reference. so we should not assume this,
if we want to use `combine()` with iterators whose dereference
return a, for instance, rvalue.
in this change, we use `std::iter_const_reference_t` instead. this
type is deduced from the behavior of the iterator instead of hardwire
it to a reference type. this allows us to use a C++20 forward_iterator
with this generic function.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
The stream_weight for repair migration is set to 2, because it requires
more work than just moving the tablet around. The stream_weight for all
other migrations are set to 1.
For historic reasons, we have (in bytes.hh) a type sstring_view which is an alias for std::string_view - since the same standard type can hold a pointer into both a seastar::sstring and std::string.
This alias in unnecessary and misleading to new developers, who might be misled to believe it is assume it is somehow different from std::string_view - when it isn't.
This series removes all uses of sstring_view (changing them to use std::string_view), and in the last patch removes the alias itself. A few functions whose name referred to "sstring" but take a std::string_view were renamed.
The patches are fairly mechanical and trivial, with no functional changes intended. To ease the review the series was split to a few smaller patches that modify specific areas of the code.
Fixes#4062.
Closesscylladb/scylladb#21617
* github.com:scylladb/scylladb:
bytes: remove unused alias sstring_view
change remaining sstring_view to std::string_view
test: change sstring_view to std::string_view
cql3: change sstring_view to std::string_view
alternator: change sstring_view to std::string_view
type: change from_sstring() to from_string_view()
cross-tree: change to_sstring_view() to to_string_view()
Wean the mutation code (at least the headers) from boost ranges to std ranges,
in order to reduce the dependency load.
Cleanup, so no backport.
Closesscylladb/scylladb#21601
* github.com:scylladb/scylladb:
partition_snapshot_row_cursor.hh: switch from boost ranges to std ranges
mutation: mutation_partition_v2.hh: switch from boost ranges to std ranges
mutation: mutation_partition.hh: switch from boost ranges to std ranges
partition_snapshot_reader.hh: drop unused include boost/range/algorithm/heap_algorithm.hpp
This change adds support for PER PARTITION LIMIT for aggregate queries.
result_set_builder gets two new functions handling partition start and
end:
- accept_partition_end for notifying that a partition has been finished.
This is also called when a page ends, so we cannot simply flush here,
as a naive implementation could do.
- accept_new_partition, where we flush_selectors() if it's indeed a new
partition (and not a continuation of the previous) and the query has a
grouping: we don't want to flush on new partition in a query like
SELECT COUNT(*) FROM foo;
select_statement::get_limit is used to evaluate the LIMIT value for both
LIMIT and PER PARTITION LIMIT. This change fixes the error message for
incorrect values passed by the user.
Currently, task_manager_module::is_aborted checks whether a task
with given id was aborted on this shard.
In tablet_repair_task_impl::run, is_aborted method is called on all
shards to check if the parent task was aborted. However, even
for aborted parent, is_aborted will return true only on owner shard
of the parent.
Pass shard param to task_manager_module::is_aborted that indicates
which shard to check.
