Do not merge tablets if that would drop the tablet_count
below the minimum provided by hints.
Split tablets if the current tablet_count is less than
the minimum tablet count calculated using the table's tablet options.
TODO: override min_tablet_count if the tablet count per shard
is greater than the maximum allowed. In this case
the tables tablet counts should be scaled down proportionally.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Rather than target_max_tablet_size. We need both the target
as well as max and min tablet sizes, so there is no sense
in keeping the max and deriving the target and the minimum
for the max value.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Use the keyspace initial_tablets for min_tablet_count, if the latter
isn't set, then take the maximum of the option-based tablet counts:
- min_tablet_count
- and expected_data_size_in_gb / target_tablet_size
- min_per_shard_tablet_count (via
calculate_initial_tablets_from_topology)
If none of the hints produce a positive tablet_count,
fall back to calculate_initial_tablets_from_topology * initial_scale.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Current implementation is inefficient as it calls
get_datacenter_token_owners_ips and then find_node(ep)
while for_each_node easily provides a host_id for
is_normal_token_owner.
Then, since we're interested only in datacenters
configure with a replication factor (but it still might be 0),
simply iterate over the dc->rf map.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Currently, if a datacenter has no replication_factor option
we consider its replication factor to be 1 in
calculate_initial_tablets_from_topology, but since
we're not going to have any replica on it, it should
be 0.
This is very minor since in the worst case, it
will pessimize the calculation and calculate a value
for initial_tablets that's higher than it could be.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Keyspace `initial` tablets option is deprecated
and may be removed in the future.
Rather than relying on `initial`:0 to always enabled
tablets, explicitly print "enabled":true when tablets
are enabled and initial_tablets=0, same as keyspace_metadata::describe.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Per-table hints should be used instead.
Note: the warning is produced by check_against_restricted_replication_strategies
which is called also from alter_keyspace_statement.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Test specifying of per-table tablet options on table creation
and alter table.
Also, add a negative test for atempting to use tablet options
with vnodes (that should fail).
And add a basic test for testing tablet options also with
materialized views.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Unlike with vnodes, each tablet is served only by a single
shard, and it is associated with a memtable that, when
flushed, it creates sstables which token-range is confined
to the tablet owning them.
On one hand, this allows for far better agility and elasticity
since migration of tablets between nodes or shards does not
require rewriting most if not all of the sstables, as required
with vnodes (at the cleanup phase).
Having too few tablets might limit performance due not
being served by all shards or by imbalance between shards
caused by quantization. The number of tabelts per table has to be
a power of 2 with the current design, and when divided by the
number of shards, some shards will serve N tablets, while others
may serve N+1, and when N is small N+1/N may be significantly
larger than 1. For example, with N=1, some shards will serve
2 tablet replicas and some will serve only 1, causing an imbalance
of 100%.
Now, simply allocating a lot more tablets for each table may
theoretically address this problem, but practically:
a. Each tablet has memory overhead and having too many tablets
in the system with many tables and many tablets for each of them
may overwhelm the system's and cause out-of-memory errors.
b. Too-small tablets cause a proliferation of small sstables
that are less efficient to acces, have higher metadata overhead
(due to per-sstable overhead), and might exhaust the system's
open file-descriptors limitations.
The options introduced in this change can help the user tune
the system in two ways:
1. Sizing the table to prevent unnecessary tablet splits
and migrations. This can be done when the table is created,
or later on, using ALTER TABLE.
2. Controlling min_per_shard_tablet_count to improve
tablet balancing, for hot tables.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
For example, nodes which are being decommissioned should not be
consider as available capacity for new tables. We don't allocate
tablets on such nodes.
Would result in higher per-shard load then planned.
Closesscylladb/scylladb#22657
in order to reduce the external header dependency, let's switch to
the standardlized std::ranges::min_element().
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22572
This config item controls how many CPU-bound reads are allowed to run in
parallel. The effective concurrency of a single CPU core is 1, so
allowing more than one CPU-bound reads to run concurrently will just
result in time-sharing and both reads having higher latency.
However, restricting concurrency to 1 means that a CPU bound read that
takes a lot of time to complete can block other quick reads while it is
running. Increase this default setting to 2 as a compromise between not
over-using time-sharing, while not allowing such slow reads to block the
queue behind them.
Fixes: #22450Closesscylladb/scylladb#22679
One of the design goals of the Alternator test suite (test/alternator)
is that developers should be able to run the tests against some already
running installation by running `cd test/alternator; pytest [--url ...]`.
Some of our presentations and documents recommend running Alternator
via docker as:
docker run --name scylla -d -p 8000:8000 scylladb/scylla:latest
--alternator-port=8000 --alternator-write-isolation=always
This only makes port 8000 available to the host - the CQL port is
blocked. We had a bug in conftest.py's get_valid_alternator_role()
which caused it to fail (and fail every single test) when CQL is
not available. What we really want is that when CQL is not available
and we can't figure out a correct secret key to connect to Alternator,
we just try a connect with a fake key - and hope that the option
alternator-enforce-authorization is turned off. In fact, this is what
the code comments claim was already happening - but we failed to
handle the case that CQL is not available at all.
After this patch, one can run Alternator with the above docker
command, and then run tests against it.
By the way, this provides another way for running any old release of
Scylla and running Alternator tests against it. We already supported
a similar feature via test/alternator/run's "--release" option, but
its implementation doesn't use docker.
Fixes#22591
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#22592
On short-pages, cut short because of a tombstone prefix.
When page-results are filtered and the filter drops some rows, the
last-position is taken from the page visitor, which does the filtering.
This means that last partition and row position will be that of the last
row the filter saw. This will not match the last position of the
replica, when the replica cut the page due to tombstones.
When fetching the next page, this means that all the tombstone suffix of
the last page, will be re-fetched. Worse still: the last position of the
next page will not match that of the saved reader left on the replica, so
the saved reader will be dropped and a new one created from scratch.
This wasted work will show up as elevated tail latencies.
Fix by always taking the last position from raw query results.
Fixes: #22620Closesscylladb/scylladb#22622
The `which` command is typically not installed on cloud OS images
and so requires the user to remember to install it (or to be prompted
by a failure to install it).
Replace it with the built-in `type` that is always there. Wrap it
in a function to make it clear what it does.
Closesscylladb/scylladb#22594
Since mid December, tests started failing with ENOMEM while
submitting I/O requests.
Logs of failed tests show IO uring was used as backend, but we
never deliberately switched to IO uring. Investigation pointed
to it happening accidentaly in commit 1bac6b75dc,
which turned on IO uring for allowing native tool in production,
and picked linux-aio backend explicitly when initializing Scylla.
But it missed that seastar-based tests would pick the default
backend, which is io_uring once enabled.
There's a reason we never made io_uring the default, which is
that it's not stable enough, and turns out we made the right
choice back then and it apparently continue to be unstable
causing flakiness in the tests.
Let's undo that accidental change in tests by explicitly
picking the linux-aio backend for seastar-based tests.
This should hopefully bring back stability.
Refs #21968.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Closesscylladb/scylladb#22695
Clean the code validating if a replication strategy can be used.
This PR consists of a bunch of unmerged https://github.com/scylladb/scylladb/pull/20088 commits - the solution to the problem that the linked PR tried to solve has been accomplished in another PR, leaving the refactor commits unmerged. The commits introduced in this PR have already been reviewed in the old PR.
No need to backport, it's just a refactor.
Closesscylladb/scylladb#22516
* github.com:scylladb/scylladb:
cql: restore validating replication strategies options
cql: change validating NetworkTopologyStrategy tags to internal_error
cql: inline abstract_replication_strategy::validate_replication_strategy
cql: clean redundant code validating replication strategy options
Currently, the session ID under which the truncate for tablets request is
running is created during the request creation and queuing. This is a problem
because this could overwrite the session ID of any ongoing operation on
system.topology#session
This change moves the creation of the session ID for truncate from the request
creation to the request handling.
Fixes#22613Closesscylladb/scylladb#22615
with_permit() creates a permit, with a self-reference, to avoid
attaching a continuation to the permit's run function. This
self-reference is used to keep the permit alive, until the execution
loop processes it. This self reference has to be carefully cleared on
error-paths, otherwise the permit will become a zombie, effectively
leaking memory.
Instead of trying to handle all loose ends, get rid of this
self-reference altogether: ask caller to provide a place to save the
permit, where it will survive until the end of the call. This makes the
call-site a little bit less nice, but it gets rid of a whole class of
possible bugs.
Fixes: #22588Closesscylladb/scylladb#22624
As of right now, materialized views (and consequently secondary
indexes), lwt and counters are unsupported or experimental with tablets.
Since by defaults tablets are enabled, training cases using those
features are currently broken.
The right thing to do here is to disable tablets in those cases.
Fixes https://github.com/scylladb/scylladb/issues/22638Closesscylladb/scylladb#22661
`validate_options` needs to be extended with
`topology` parameter, because NetworkTopologyStrategy needs to validate if every
explicitly listed DC is really existing. I did cut corner a bit and
trimmed the message thrown when it's not the case, just to avoid passing
and extra parameter (ks name) to the `validate_options`
function, as I find the longer message to be a bit redundant (the driver will
receive info which KS modification failed).
The tests that have been commented out in the previous commit have been
restored.
The check for `replication_factor` tag in
`network_topology_strategy::validate_options` is redundant for 2 reasons:
- before we reach this part of the code, the `replication_factor` tag
is replaced with specific DC names
- we actually do allow for `replication_factor` tag in
NetworkTopologyStrategy for keyspaces that have tablets disabled.
This code is unreachable, hence changing it to an internal error, which
means this situation should never occur.
The place that unrolls `replication_factor` tag checked for presence of
this tag ignoring the case, which lead to an unexpected behaviour:
- `replication_factor` tag (note the lowercase) was unrolled, as
explained above,
- the same tag but written in any other case resulted in throwing a vague
message: "replication_factor is an option for SimpleStrategy, not
NetworkTopologyStrategy".
So we're changing this validation to accept and unroll only the
lowercase version of this tag. We can't ignore the case here, as this
tag is present inside a json, and json is case-sensitive, even though the
CQL itself is case insensitive.
Added a test that passes for both scylla and cassandra.
Fixes: #15336
task_stats contains short info about a task. To get a list of task_stats
in the module, one needs to request /task_manager/list_module_tasks/{module}.
To make identification and navigation between tasks easier, extend
task_stats to contain shard, start_time, and end_time.
Closesscylladb/scylladb#22351
tablet_repair_task_impl is run as a part of tablet repair. Make it
a child of tablet repair virtual task.
tablet_repair_task_impl started by /storage_service/repair_async API
(vnode repair) does not have a parent, as it is the top-level task
in that case.
No backport needed; new functionality
Closesscylladb/scylladb#22372
* github.com:scylladb/scylladb:
test: add test to check tablet repair child
service: add child for tablet repair virtual task
Currently, when the tablet repair is started, info regarding
the operation is kept in the system.tablets. The new tablet states
are reflected in memory after load_topology_state is called.
Before that, the data in the table and the memory aren't consistent.
To check the supported operations, tablet_virtual_task uses in-memory
tablet_metadata. Hence, it may not see the operation, even though
its info is already kept in system.tablets table.
Run read barrier in tablet_virtual_task::contains to ensure it will
see the latest data. Add a test to check it.
Fixes: #21975.
Closesscylladb/scylladb#21995
This was originally an attempt to reduce the compile time of this
translation unit, but apparently it doesn't work. Still, it has
the effect of converting stack traces that say "set_storage_service"
and refer to some lambda to stack traces that refer to the operation
being performed, so it's a net positive.
To faciliate the change, we introduce new functions rest_bind(),
similar to (and in fact wrapping) std::bind_front(), that capture
references like the lambdas did originally. We can't use
std::bind_front directly since the call to
seastar::httpd::path_description::set() cannot be disambiguated
after the function is obscured by the template returned by
std::bind_front. The new function rest_bind() has constraints
to understand which overload is in use.
Closesscylladb/scylladb#22526
This PR enhances the internode_compression configuration option in two ways:
1. Add validation for option values
Previously, we silently defaulted to 'none' when given invalid values. Now we
explicitly validate against the three supported values (all, dc, none) and
reject invalid inputs. This provides better error messages when users
misconfigure the option.
2. Fix documentation rendering
The help text for this option previously used C++ escape sequences which
rendered incorrectly in Sphinx-generated HTML. We now use bullet points with
'*' prefix to list the available values, matching our documentation style
for other config options. This ensures consistent rendering in both CLI
and HTML outputs.
Note: The current documentation format puts type/default/liveness information
in the same bullet list as option values. This affects other config options
as well and will need to be addressed in a separate change.
---
this improves the handling of invalid option values, and improves the doc rendering, neither of which is critical. hence no need to backport.
Closesscylladb/scylladb#22548
* github.com:scylladb/scylladb:
config: validate internode_compression option values
config: start available options with '*'
Bug https://bugs.python.org/issue26789 is resolved in python 3.10.
The frozen tool chain uses python 3.12. Since this is a supported and
recommended way for work environment, removing workaround and bumping
requirements for a newer python version.
Closesscylladb/scylladb#22627
Following the work done in ed4bfad5c3, the action is failing with the
following error:
```
Error: Input required and not supplied: token
```
It is due ot missing permissions in the workflow, adding it
Closesscylladb/scylladb#22630
tablet_repair_task_impl is run as a part of tablet repair. Make it
a child of tablet repair virtual task.
tablet_repair_task_impl started by /storage_service/repair_async API
(vnode repair) does not have a parent, as it is the top-level task
in that case.
* seastar 71036ebcc0...5b95d1d798 (3):
> rpc stream: do not abort stream queue if stream connection was closed without error
> resource: fallback to sysconf when failed to detect memory size from hwloc
> Merge 'scheduling_group: improve scheduling group creation exception safety' from Michael Litvak
scylla-gdb.py adjusted for scheduling_group_specific data structure
changes in Seastar. As part of that, a gratuitous dereference of
std::unique_ptr, which fails for std::unique_ptr<void*, ...>, was
removed.
The test expects and asserts that after wait_for_view is completed we
read the view_build_status table and get a row for each node and view.
But this is wrong because wait_for_view may have read the table on one
node, and then we query the table on a different node that didn't insert
all the rows yet, so the assert could fail.
To fix it we change the test to retry and check that eventually all
expected rows are found and then eventually removed on the same host.
Fixesscylladb/scylladb#22547Closesscylladb/scylladb#22585
The view builder builds a view by going over the entire token ring,
consuming the base table partitions, and generating view updates for
each partition.
A view is considered as built when we complete a full cycle of the
token ring. Suppose we start to build a view at a token F. We will
consume all partitions with tokens starting at F until the maximum
token, then go back to the minimum token and consume all partitions
until F, and then we detect that we pass F and complete building the
view. This happens in the view builder consumer in
`check_for_built_views`.
The problem is that we check if we pass the first token F with the
condition `_step.current_token() >= it->first_token` whenever we consume
a new partition or the current_token goes back to the minimum token.
But suppose that we don't have any partitions with a token greater than
or equal to the first token (this could happen if the partition with
token F was moved to another node for example), then this condition will never be
satisfied, and we don't detect correctly when we pass F. Instead, we
go back to the minimum token, building the same token ranges again,
in a possibly infinite loop.
To fix this we add another step when reaching the end of the reader's
stream. When this happens it means we don't have any more fragments to
consume until the end of the range, so we advance the current_token to
the end of the range, simulating a partition, and check for built views
in that range.
Fixesscylladb/scylladb#21829Closesscylladb/scylladb#22493
Add two cqlpy tests that reproduce a bug where a secondary index query
returns more rows than the specified limit. This occurs when the indexed
column is a partition key column or the first clustering key column,
the query result spans multiple partitions, and the last partition
causes the limit to be exceeded.
`test/cqlpy/run --release ...` shows that the tests fail for Scylla
versions all the way back to 4.4.0. Older Scylla versions fail with a
syntax error in CQL query which suggests some incompatibility in the
CQL protocol. That said, this bug is not a regression.
The tests pass in Cassandra 5.0.2.
Refs #22158.
Signed-off-by: Nikos Dragazis <nikolaos.dragazis@scylladb.com>
Closesscylladb/scylladb#22513
std::any_of was included by C++11, and boost::algorithm::any_of() is
provided by Boost for users stuck in the pre-C++11 era. in our case,
we've moved into C++23, where the ranges variant of this algorithm
is available.
in order to reduce the header dependency, let's switch to
`std::ranges::any_of()`.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22503
Materialized views with tablets are not stable yet, but we want
them available as an experimental feature, mainly for teseting.
The feature was added in scylladb/scylladb#21833,
but currently it has no effect. All tests have been updated to use the
feature, so we should finally make it work.
This patch prevents users from creating materialized views in keyspaces
using tablets when the VIEWS_WITH_TABLETS feature is not enabled - such
requests will now get rejected.
Fixesscylladb/scylladb#21832Closesscylladb/scylladb#22217
