After this series, tablet replication can handle the scenario of bootstrapping new nodes. The ownership is distributed indirectly by the means of a load-balancer which moves tablets around in the background. See docs/dev/topology-over-raft.md for details.
The implementation is by no means meant to be perfect, especially in terms of performance, and will be improved incrementally.
The load balancer will be also kicked by schema changes, so that allocation/deallocation done during table creation/drop will be rebalanced.
Tablet data is streamed using existing `range_streamer`, which is the infrastructure for "the old streaming". This will be later replaced by sstable transfer once integration of tablets with compaction groups is finished. Also, cleanup is not wired yet, also blocked by compaction group integration.
Closes#14601
* github.com:scylladb/scylladb:
tests: test_tablets: Add test for bootstraping a node
storage_service: topology_coordinator: Implement tablet migration state machine
tablets: Introduce tablet_mutation_builder
service: tablet_allocator: Introduce tablet load balancer
tablets: Introduce tablet_map::for_each_tablet()
topology: Introduce get_node()
token_metadata: Add non-const getter of tablet_metadata
storage_service: Notify topology state machine after applying schema change
storage_service: Implement stream_tablet RPC
tablets: Introduce global_tablet_id
stream_transfer_task, multishard_writer: Work with table sharder
tablets: Turn tablet_id into a struct
db: Do not create per-keyspace erm for tablet-based tables
tablets: effective_replication_map: Take transition stage into account when computing replicas
tablets: Store "stage" in transition info
doc: Document tablet migration state machine and load balancer
locator: erm: Make get_endpoints_for_reading() always return read replicas
storage_service: topology_coordinator: Sleep on failure between retries
storage_service: topology_coordinator: Simplify coordinator loop
main: Require experimental raft to enable tablets
The method is called by db::truncate_table_on_all_shards(), its call-chain, in turn, starts from
- proxy::remote::handle_truncate()
- schema_tables::merge_schema()
- legacy_schema_migrator
- tests
All of the above are easy to get system_keyspace reference from. This, in turn, allows making the method non-static and use query_processor reference from system_keyspace object in stead of global qctx
Closes#14778
* github.com:scylladb/scylladb:
system_keyspace: Make save_truncation_record() non-static
code: Pass sharded<db::system_keyspace>& to database::truncate()
db: Add sharded<system_keyspace>& to legacy_schema_migrator
The function dispatch a background operation that must be
waited on in stop().
Fixes scylladb/scylladb#14791
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Closes#14797
Identifies tablet in the scope of the whole cluster. Not to be
confused with tablet replicas, which all share global_tablet_id.
Will be needed by load balancer and tablet migration algorithm to
identify tablets globally.
This erm is not updated when replicating token metadata in
storage_service::replicate_to_all_cores() so will pin token metadata
version and prevent token metadata barrier from finishing.
It is not necessary to have per-keyspace erm for tablet-based tables,
so just don't create it.
It's needed to implement tablet migration. It stores the current step
of tablet migration state machine. The state machine will be advanced
by the topology change coordinator.
See the "Tablet migration" section of topology-over-raft.md
Just a simplification.
Drop the test case from token_metadata which creates pending endpoints
without normal tokens. It fails after this change with exception:
"sorted_tokens is empty in first_token_index!" thrown from
token_metadata::first_token_index(), which is used when calculating
normal endpoints. This test case is not valid, first node inserts
its tokens as normal without going through bootstrap procedure.
When messaging_service shuts down it first sets _shutting_down to true
and proceeds with stopping clients and servers. Stopping clients, in
turn, is calling client.stop() on each.
Setting _shutting_down is used in two places.
First, when a client is stopped it may happen that it's in the middle of
some operation, which may result in call to remove_error_rpc_client()
and not to call .stop() for the second time it just does nothing if the
shutdown flag is set (see 357c91a076).
Second, get_rpc_client() asserts that this flag is not set, so once
shutdown started it can make sure that it will call .stop() on _all_
clients and no new ones would appear in parallel.
However, after shutdown() is complete the _clients vector of maps
remains intact even though all clients from it are stopped. This is not
very debugging-friendly, the clients are better be removed on shutdown.
fixes: #14624
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closes#14632
It is quite common to stop a tested scylla process with ^C, which will
raise KeyboardInterrupt from subprocess.run(). Catch and swallow this
exception, allowing the post-processing to continue.
The interrupted process has to handle the interrupt correctly too --
flush the coverage data even on premature exit -- but this is for
another patch.
Closes#14815
The new test detected a stack-use-after-return when using table's
as_mutation_source_excluding_staging() for range reads.
This doesn't really affect view updates that generate single
key reads only. So the problem was only stressed in the recently
added test. Otherwise, we'd have seen it when running dtests
(in debug mode) that stress the view update path from staging.
The problem happens because the closure was feeded into
a noncopyable_function that was taken by reference. For range
reads, we defer before subsequent usage of the predicate.
For single key reads, we only defer after finished using
the predicate.
Fix is about using sstable_predicate type, so there won't
be a need to construct a temporary object on stack.
Fixes#14812.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Closes#14813
in this series, we use different table names in simple_backlog_controller_test. this test is a test exercising sstables compaction strategies. and it creates and keeps multiple tables in a single test session. but we are going to add metrics on per-table basis, and will use the table's ks and cf as the counter's labels. as the metrics subsystem does not allow multiple counters to share the same label. the test will fail when the metrics are being added.
to address this problem, in this change
1. a new ctor is added for `simple_schema`, so we can create `simple_schema` with different names
2. use the new ctor in simple_backlog_controller_test
Fixes#14767Closes#14783
* github.com:scylladb/scylladb:
test: use different table names in simple_backlog_controller_test
test/lib/simple_schema: add ctor for customizing ks.cf
test/lib/simple_schema: do not hardwire ks.cf
This commit adds the information on how to install ScyllaDB
without root privileges (with "unified installer", but we've
decided to drop that name - see the page title).
The content taken from the website
https://www.scylladb.com/download/?platform=tar&version=scylla-5.2#open-source
is divided into two sections: "Download and Install" and
"Configure and Run ScyllaDB".
In addition, the "Next Steps" section is also copied from
the website, and adjusted to be in sync with other installation
pages in the docs.
Refs https://github.com/scylladb/scylla-docs/issues/4091Closes#14781
In this commit we just pass a fencing_token
through hint_mutation RPC verb.
The hints manager uses either
storage_proxy::send_hint_to_all_replicas or
storage_proxy::send_hint_to_endpoint to send a hint.
Both methods capture the current erm and use the
corresponding fencing token from it in the
mutation or hint_mutation RPC verb. If these
verbs are fenced out, the server stale_topology_exception
is translated to a mutation_write_failure_exception
on the client with an appropriate error message.
The hint manager will attempt to resend the failed
hint from the commitlog segment after a delay.
However, if delivery is unsuccessful, the hint will
be discarded after gc_grace_seconds.
Closes#14580
We don't load gossiper endpoint states in `storage_service::join_cluster` if `_raft_topology_change_enabled`, but gossiper is still needed even in case of `_raft_topology_change_enabled` mode, since it still contains part of the cluster state. To work correctly, the gossiper needs to know the current endpoints. We cannot rely on seeds alone, since it is not guaranteed that seeds will be up to date and reachable at the time of restart.
The problem was demonstrated by the test `test_joining_old_node_fails`, it fails occasionally with `experimental_features: [consistent-topology-changes]` on the line where it waits for `TEST_ONLY_FEATURE` to become enabled on all nodes. This doesn't happen since `SUPPORTED_FEATURES` gossiper state is not disseminated, and feature_service still relies on gossiper to disseminate information around the cluster.
The series also contains a fix for a problem in `gossiper::do_send_ack2_msg`, see commit message for details.
Fixes#14675Closes#14775
* github.com:scylladb/scylladb:
storage_service: restore gossiper endpoints on topology_state_load fix
gossiper: do_send_ack2_msg fix
If semaphore mismatch occurs, check whether both semaphores belong
to user. If so, log a warning, log a `querier_cache_scheduling_group_mismatches` stat and drop cached reader instead of throwing an error.
Until now, semaphore mismatch was only checked in multi-partition queries. The PR pushes the check to `querier_cache` and perform it on all `lookup_*_querier` methods.
The mismatch can happen if user's scheduling group changed during
a query. We don't want to throw an error then, but drop and reset
cached reader.
This patch doesn't solve a problem with mismatched semaphores because of changes in service levels/scheduling groups but only mitigate it.
Refers: https://github.com/scylladb/scylla-enterprise/issues/3182
Refers: https://github.com/scylladb/scylla-enterprise/issues/3050Closes: #14770Closes#14736
* github.com:scylladb/scylladb:
querier_cache: add stats of scheduling group mismatches
querier_cache: check semaphore mismatch during querier lookup
querier_cache: add reference to `replica::database::is_user_semaphore()`
replica:database: add method to determine if semaphore is user one
We don't load gossiper endpoint states in
storage_service::join_cluster if
_raft_topology_change_enabled, but gossiper
is still needed even in case of
_raft_topology_change_enabled mode, since it
still contains part of the cluster state.
To work correctly, the gossiper needs to know
the current endpoints. We cannot rely on seeds alone,
since it is not guaranteed that seeds will be
up to date and reachable at the time of restart.
The specific scenario of the problem: cluster with
three nodes, the second has the first in seeds,
the third has the first and second. We restart all
the nodes simultaneously, the third node uses its
seeds as _endpoints_to_talk_with in the first gossiper round
and sends SYN to the first and sedond. The first node
hasn't started its gossiper yet, so handle_syn_msg
returns immediately after if (!this->is_enabled());
The third node receives ack from the second node and
no communication from the first node, so it fills
its _live_endpoints collection with the second node
and will never communicate with the first node again.
The problem was demonstrated by the test
test_joining_old_node_fails, it fails occasionally with
experimental_features: [consistent-topology-changes]
on the line where it waits for TEST_ONLY_FEATURE
to become enabled on all nodes. This doesn't happen
since SUPPORTED_FEATURES gossiper state is not
disseminated because of the problem described above.
The first commit is needed since add_saved_endpoint
adds the endpoint with some default app states with locally
incrementing versions and without that fix gossiper
refuses to fill the real app states for this endpoint later.
Fixes: #14675
Fixes#14668
In #14668, we have decided to introduce a new `scylla.yaml` variable for the schema commitlog segment size and set it to 128MB. The reason is that segment size puts a limit on the mutation size that can be written at once, and some schema mutation writes are much larger than average, as shown in #13864. This `schema_commitlog_segment_size_in_mb variable` variable is now added to `scylla.yaml` and `db/config`.
Additionally, we do not derive the commitlog sync period for schema commitlog anymore because schema commitlog runs in batch mode, so it doesn't need this parameter. It has also been discussed in #14668.
Closes#14704
* github.com:scylladb/scylladb:
replica: do not derive the commitlog sync period for schema commitlog
config: set schema_commitlog_segment_size_in_mb to 128
config: add schema_commitlog_segment_size_in_mb variable
This commit is a first part of the fix for #14675.
The issue is about the test test_joining_old_node_fails
faling occasionally with
experimental_features: [consistent-topology-changes].
The next commit contains a fix for it, here we
solve the pre-existing gossiper problem
which we stumble upon after the fix.
Local generation for addr may have been
increased since the current node sent
an initial SYN. Comparing versions across different
generations in get_state_for_version_bigger_than
could result in loosing some app states with
smaller versions.
More specifically, consider a cluster with nodes
.1, .2, .3, .3 has .1 and .2 as seeds, .2 has .1
as a seed. Suppose .2 receives a SYN from .3 before
its gossiper starts, and it has a
version 0.24 for .1 in endpoint_states.
The digest from .3 contains 0.25 as a version for .1,
so examine_gossiper produces .1->0.24 as a digest
and this digest is send to .3 as part of the ack.
Before processing this ack, .3 processed an ack from
.1 (scylla sends SYN to many nodes) and updates
its endpoint_states according to it, so now it
has .1->100500.32 for .1. Then
we get to do_send_ack2_msg and call
get_state_for_version_bigger_than(.1, 24).
This returns properties which has version > 24,
ignoring a lot of them with smaller versions
which has been received from .1. Also,
get_state_for_version_bigger_than updates
generation (it copies get_heart_beat_state from
.3), so when we apply the ack in handle_ack2_msg
at .2 we update the generation and now the
skipped app states will only be updated on .2
if somebody change them and increment their version.
Cassandra behaviour is the same in this case
(see https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/gms/GossipDigestAckVerbHandler.java#L86). This is probably less
of a problem for them since most of the time they
send only one SYN in one gossiper round
(save for unreachable nodes), so there is less
room for conflicts.
before this change, add_version_library() is a single function
which accomplishes two tasks:
1. build scylla-version target using
2. add an object library
but this has two problems:
1. we should run `SCYLLA-VERSION-GEN` at configure time, instead
of at build time. otherwise the targets which read from the
SCYLLA-{VERSION, RELEASE, PRODUCT}-FILE cannot access them,
unless they are able to read them in their build rules. but
they always use `file(STRINGS ..)` to read them, and thsee
`file()` command is executed at configure time. so, this
is a dead end.
2. we repeat the `file(STRING ..)` multiple places. this is
not ideal if we want to minimize the repeatings.
so, to address this problem, in this change:
1. use `execute_process()` instead of `add_custom_command()`
for generating these *-FILE files. so they are always ready
at build time. this partially reverts bb7d99ad37.
2. extract `generate_scylla_version()` out of `add_version_library()`.
so we can call the former much earlier than the latter.
this would allow us to reference the variables defined by
the `generate_scylla_version()` much earlier.
3. define cached strings in the extracted function, so that
they can consumed by other places.
4. reference the cached variables in `build_submodule.cmake`.
also, take this opportunity to fix the version string
used in build_submodule.cmake: we should have used
`scylla_version_tilde`.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closes#14769
Previously semaphore mismatch was checked only in multi-partition
queries and if happened, an internal error was thrown.
This commit pushed the check down to `querier_cache`, so each
`lookup_*_querier` method will check for the mismatch.
What's more, if semaphore mismatch occurs, check whether both semaphores belong
to user. If so, log a warning and drop cached reader instead of
throwing an error.
The mismatch can happen if user's scheduling group changed during
a query. We don't want to throw an error then, but drop and reset
cached reader.
Before choosing a function, we prepare the arguments that can be
prepared without a receiver. Preparing an argument makes
its type known, which allows to choose the best overload
among many possible functions.
The function that prepared the argument passes the unprepared
argument by mistake. Let's fix it so that it actually uses
the prepared argument.
Signed-off-by: Jan Ciolek <jan.ciolek@scylladb.com>
Closes#14786
We allow inserting column values using a JSON value, eg:
```cql
INSERT INTO mytable JSON '{ "\"myKey\"": 0, "value": 0}';
```
When no JSON value is specified, the query should be rejected.
Scylla used to crash in such cases. A recent change fixed the crash
(https://github.com/scylladb/scylladb/pull/14706), it now fails
on unwrapping an uninitialized value, but really it should
be rejected at the parsing stage, so let's fix the grammar so that
it doesn't allow JSON queries without JSON values.
A unit test is added to prevent regressions.
Refs: https://github.com/scylladb/scylladb/pull/14707
Fixes: https://github.com/scylladb/scylladb/issues/14709
Signed-off-by: Jan Ciolek <jan.ciolek@scylladb.com>
Closes#14785
in `simple_backlog_controller_test`, we need to have multiple tables
at the same time. but the default constructor of `simple_schema` always
creates schema with the table name of "ks.cf". we are going to have
a per-table metrics. and the new metric group will use the table name
as its counter labels, so we need to either disable this per-table
metrics or use a different table name for each table.
as in real world, we don't have multiple tables at the same time. it
would be better to stop reusing the same table name in a single test
session. so, in this change, we use a random cf_name for each of
the created table.
Fixes#14767
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
some low level tests, like the ones exercising sstables, creates
multiple tables. and we are going to add per-table metrics and
the new metrics uses the ks.cf as part of its unique id. so,
once the per-table metrics is enabled, the sstable tests would fail.
as the metrics subsystem does not allow registering multiple
metric groups with the same name.
so, in this change, we add a new constructor for `simple_schema`,
so that we can customize the the schema's ks and cf when creating
the `simple_schema`. in the next commit, we will use this new
constructor in a sstable test which creates multiple tables.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
instead, query the name of ks and cf from the scheme. this change
prepare us for the a simple_schema whose ks and cf can be customized
by its contructor.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
The arguments goes via the db::(drop|truncate)_table_on_all_shards()
pair of calls that start from
- storage_proxy::remote: has its sys.ks reference already
- schema_tables::merge_schema: has sys.ks argument already
- legacy_schema_migrator: the reference was added by previous patch
- tests: run in cql_test_env with sys.ks on board
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
One of the class' methods calls db::drop_table_on_all_shards() that will
need sys.ks. in the next patch.
The reference in question is provided from the only caller -- main.cc
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
in 46616712, we tried to keep the tmpdir only if the test failed,
and keep up to 1 of them using the recently introduced
option of `tmp_path_retention_count`. but it turns out this option
is not supported by the pytest used by our jenkins nodes, where we
have pytest 6.2.5. this is the one shipped along with fedora 36.
so, in this change, the tempdir is removed if the test completes
without failures. as the tempdir contains huge number of files,
and jenkins is quite slow scanning them. after nuking the tempdir,
jenkins will be much faster when scanning for the artifacts.
Fixes#14690
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closes#14772
