When applying a counter mutation, use apply_on_shards to apply the
mutation on all write shards, similarly to the way other mutations are
applied in the storage proxy. Previously the mutation was applied only
on the current shard which is the read shard.
This is needed to respect the write_both stages of intranode migration
where we need to apply the mutation on both the old and the new shards.
Refactor the counter update to split the functions and have them called
by the storage proxy to prepare for a later change.
Previously in mutate_counter the storage proxy calls the replica
function apply_counter_update that does a few things:
1. checks that the operation can be done: check timeout, disk utilization
2. acquire counter locks
3. do read-modify-write and transform the counter mutation
4. apply the mutation in the replica
In this commit we change it so that these functions are split and called
from the storage proxy, so that we have better control from the storage
proxy when we change it later to work across multiple shards. For
example, we will want to acquire locks on multiple shards, transform it
on one shard, and then apply the mutation on multiple shards.
After the change it works as follows in storage proxy:
1. acquire counter locks
2. call replica prepare to check the operation and transform the mutation
3. call replica apply to apply the transformed mutation
Add a RAII guard for counter update that holds the counter locks and the
table operation, and extract the creation of the guard to a separate
function.
This prepares it for a later change where we will want to obtain the
guard externally from the storage proxy.
The compaction manager backlog is exposed via metrics, but if static
shares are set, the backlog is never calculated. As a result, there is
no way to determine the backlog and if the static shares need
adjustment. Fix that by calculating backlog even when static shares are
set.
Fixes#26287
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
Closesscylladb/scylladb#26778
Before this patch, when a base table has many materialized views,
each write to this table can start up to 128 view updates in parallel.
With high client write concurrency, the actual concurrency of writes
executed on the node may grow unexpectedly, which can lead to higher
latency and higher memory usage compared to a sequential approach.
In this patch we add a per-shard, per-service-level semaphore which
limits the number of concurrent view updates processed on the shard
in this service level to a constant value. We take one unit from the
semaphore for each local view update write, and releasing it when it
finishes. The remote view updates do not take units from the semaphore
because they don't consume nearly as much processing power and they
are limited by another semaphore based on their memory usage.
Fixes https://github.com/scylladb/scylladb/issues/25341Closesscylladb/scylladb#25456
* github.com:scylladb/scylladb:
mv: limit concurrent view updates from all sources
database: rename _view_update_concurrency_sem to _view_update_memory_sem
Before this patch, when a base table has many materialized views,
each write to this table can start up to 128 view updates in parallel.
With high client write concurrency, the actual concurrency of writes
executed on the node may grow unexpectedly, which can lead to higher
latency and higher memory usage compared to a sequential approach.
In this patch we add a per-shard, per-service-level semaphore which
limits the number of concurrent view updates processed on the shard
in this service level to a constant value. We take one unit from the
semaphore for each local view update write, and releasing it when it
finishes. The remote view updates do not take units from the semaphore
because they don't consume nearly as much processing power and they
are limited by another semaphore based on their memory usage.
The effect of this patch can also be observed when writing to a base
table with a large number of materialized views, like in the
materialized_views_test.py::TestMaterializedViews::test_many_mv_concurrent
dtest. In that test, if we perform a full scan in parallel to a write
workload with a concurrency of 100 to a table with 100 views, the scan
would sometimes timeout because it would effectively get 1/10000 of cpu.
With this patch, the cpu concurrency of view updates was limited to 128
(we ran both writes and scan in the same service level), and the scan
no longer timed out.
Fixes https://github.com/scylladb/scylladb/issues/25341
Problems addressed by this PR
* Missing barrier before cleanup: If a node was bootstrapped before cleanup, some request coordinators could still be in `write_both_read_new` and send stale requests to replicas being cleaned up.
* Sessions not drained before cleanup: We lacked protection against stale streaming or repair operations.
* `sstable_vnodes_cleanup_fiber()` calling `flush_all_tables()` under group0 lock: This caused SCT test failures (see [this comment](https://github.com/scylladb/scylladb/issues/25333#issuecomment-3298859046) for details).
* Issues with `storage_proxy::start_write()` used by `sstable_vnodes_cleanup_fiber`:
* The result of `start_write()` was not held during `abstract_write_response_handler::apply_locally`, so coordinator-local writes were not properly awaited.
* Synchronization was racy — `start_write()` was not atomic with the fence check, allowing stale writes to sneak in if `fence_version` changed in between.
* It waited for all writes, including local tables and tablet-based tables, which is redundant because `sstable_vnodes_cleanup_fiber` does not apply to them.
* It also waited for writes with versions greater than the current `fence_version`, which is unnecessary.
Fixesscylladb/scylladb#26150
backport: this PR fixes several issues with the vnodes cleanup procedure, but it doesn't seem they are critical enough to deserve backporting
Closesscylladb/scylladb#26315
* https://github.com/scylladb/scylladb:
test_automatic_cleanup: add test_cleanup_waits_for_stale_writes
test_fencing: fix due to new version increment
test_automatic_cleanup: clean it up
storage_proxy: wait for closing sessions in sstable cleanup fiber
storage_proxy: rename await_pending_writes -> await_stale_pending_writes
storage_proxy: use run_fenceable_write
storage_proxy: abstract_write_response_handler: apply_locally: extract post fence check
storage_proxy: introduce run_fenceable_write
storage_proxy: move update_fence_version from shared_token_metadata
storage_proxy: fix start_write() operation scope in apply_locally
storage_proxy: move post fence check into handle_write
storage_proxy: move fencing into mutate_counter_on_leader_and_replicate
storage_proxy::handle_read: add fence check before get_schema
storage_service: rebrand cleanup_fiber to vnodes_cleanup_fiber
sstable_cleanup_fiber: use coroutine::parallel_for_each
storage_service: sstable_cleanup_fiber: move flush_all_tables out of the group0 lock
topology_coordinator: barrier before cleanup
topology_coordinator: small start_cleanup refactoring
global_token_metadata_barrier: add fenced flag
`select * from mutation_fragment()` queries don't return partitions which are completely empty or only contain tombstones which are all garbage collectible. This is because the underlying `mutation_dump` mechanism has a separate query to discover partitions for scans. This query is a regular mutation scan, which is subject to query compaction and garbage collection. Disable the query compaction for mutation queries executed on behalf of mutation fragment queries, so *all* data is visible in the result, even that which is fully garbage collectible.
Fixesscylladb/scylladb#23707.
Scans for mutation-fragment are very rare, so a backport is not necessary. We can backport on-demand.
Closesscylladb/scylladb#26227
* github.com:scylladb/scylladb:
replica/mutation_dump: multi_range_partition_generator: disable garbage-collection
replica: add tombstone_gc_enabled parameter to mutation query methods
mutation/mutation_compactor: remove _can_gc member
tombstone_gc: add tombstone_gc_state factory methods for gc_all and no_gc
In the following commit, we'll introduce a new semaphore for view updates
that limits their concurrency by view update count. To avoid confusion,
we rename the existing semaphore that tracks the memory used by concurrent
view updates and related objects accordingly.
This is a follow-up of the previous fix: https://github.com/scylladb/scylladb/pull/26030
The test test_user_writes_rejection starts a 3-node cluster and
creates a large file on one of the nodes, to trigger the out-of-space
prevention mechanism, which should reject writes on that node.
It waits for the log message 'Setting critical disk utilization mode: true'
and then executes a write expecting the node to reject it.
Currently, the message is logged before the `_critical_disk_utilization`
variable is actually updated. This causes the test to fail sporadically
if it runs quickly enough.
The fix splits the logging into two steps:
1. "Asked to set critical disk utilization mode" - logged before any action
2) "Set critical disk utilization mode" - logged after `_critical_disk_utilization` has been updated
The tests are updated to wait for the second message.
Fixes https://github.com/scylladb/scylladb/issues/26004Closesscylladb/scylladb#26392
This patch series introduces several tests that check number of exceptions that happens during various replica operations. The goal is to have a set of tests that can catch situations where number of exceptions per operation increases. It makes exception throw regressions easier to catch.
The tests cover apply counter update and apply functionalities in the database layer.
There are more paths that can be checked, like various semaphore wait timeouts located deeper in the code. This set of tests does not cover all code paths.
Fixes#18164
This is an improvement. No backport needed.
Closesscylladb/scylladb#25992
* github.com:scylladb/scylladb:
test: cluster: test replica write timeout
database: parameterize apply_counter_update_delay_5s injector value
test: cluster: test replica exceptions - test rate limit exceptions
This patch introduces test `test_replica_database_apply_timeout`.
It tests timeout on database write. The test uses error injection
that returns timeout error if the injection `database_apply_force_timeout`
is enabled.
Refs #18164
Parameterize `apply_counter_update_delay_5s` injector value. Instead of
sleeping 5s when the injection is active, read parameter value that
specifies sleep duration. To reflect these changes, it is renamed to
`apply_counter_update_delay_ms` and the sleep duration is specified in
milliseconds.
Refs #18164
The series adds an experimental flag for strongly consistent tables and extends "CREATE KEYSPACE" ddl with `consistency` option that allows specifying the consistency mode for the keyspace.
Closesscylladb/scylladb#26116
* github.com:scylladb/scylladb:
schema: Allow configuring consistency setting for a keyspace
db: experimental consistent-tablets option
We want to add strongly consistent tables as an option. We will have
two kind of strongly consistent tables: globally consistent and locally
consistent. The former means that requests from all DCs will be globally
linearisable while the later - only requests to the same DCs will be
linearisable. To allow configuring all the possibilities the patch
adds new parameter to a keyspace definition "consistency" that can be
configured to be `eventual`, `global` or `local`. Non eventual setting
is supported for tablets enabled keyspaces only. Since we want to start
with implementing local consistency configuring global consistency will
result in an error for now.
Allow disabling tombstone gc on a per-query basis for mutation queries.
This is achieved by a bool flag passed to mutation query variants like
`query_mutations_on_all_shards()` and `database::mutation_query()`,
which is then propagated down to compaction_mutation_state.
The future user (in the next patch) is the SELECT * FROM
MUTATION_FRAGMENTS() statement which wants to see dead partitions
(and rows) when scanning a table. Currently, due to garbage collections,
said statement can miss partitions which only contain
garbage-collectible tombstones.
This change extends the CQL replication options syntax so the replication factor can be stated as a list of rack names.
For example: { 'mydatacenter': [ 'myrack1', 'myrack2', 'myrack4' ] }
Rack-list based RF can coexist with the old numerical RF, even in the same keyspace for different DCs.
Specifying the rack list also allows to add replicas on the specified racks (increasing the replication factor), or decommissioning certain racks from their replicas (by omitting them from the current datacenter rack-list). This will allow us to keep the keyspace rf-rack-valid, maintaining guarantees, while allowing adding/removing racks. In particular, this will allow us to add a new DC, which happens by incrementally increasing RF in that DC to cover existing racks.
Migration from numerical RF to rack-list is not supported yet. Migration from rack-list to numerical RF is not planned to be supported.
New feature, no backport required.
Co-authored with @bhalevy
Fixes https://github.com/scylladb/scylladb/issues/25269
Fixes https://github.com/scylladb/scylladb/issues/23525Closesscylladb/scylladb#26358
* github.com:scylladb/scylladb:
tablets: load_balancer: Recognize that tablets are confined to racks when computing desired tablet count
locator: Make hasher for endpoint_dc_rack globally accessible
test: tablets: Add test for replica allocation on rack list changes
test: lib: topology_builder: generate unique rack names
test: Add tests for rack list RF
doc: Document rack-list replication factor
topology_coordinator: Restore formatting
topology_coordinator: Cancel keyspace alter on broader set of errors
topology_coordinator: Make keyspace alter process options through as_ks_metadata_update()
cql3: ks_prop_defs: Preserve old options
cql3: ks_prop_defs: Introduce flattened()
locator: Recognize rack list RF as valid in assert_rf_rack_valid_keyspace()
tablet_allocator: Respect binding replicas to racks
locator: network_topology_strategy: Respect rack list when reallocating tablets
cql3: ks_prop_defs: Fail with more information when options are not in expected format
locator, cql3: Support rack lists in replication options
cql3: Fail early on vnode/tablet flavor alter
cql3: Extract convert_property_map() out of Cql.g
schema: Use definition from the header instead of open-coding it
locator: Abstract obtaining the number of replicas from replication_strategy_config_option
cql3, locator: Use type aliases for option maps
locator: Add debug logging
locator: Pass topology to replication strategy constructor
abstract_replication_strategy, network_topology_strategy: add replication_factor_data class
Materialized views are currently in the experimental phase and using them
in tablet-based keyspaces requires starting Scylla with an experimental feature,
`views-with-tablets`. Any attempts to create a materialized view or secondary
index when it's not enabled will fail with an appropriate error.
After considerable effort, we're drawing close to bringing views out of the
experimental phase, and the experimental feature will no longer be needed.
However, materialized views in tablet-based keyspaces will still be restricted,
and creating them will only be possible after enabling the configuration option
`rf_rack_valid_keyspaces`. That's what we do in this PR.
In this patch, we adjust existing tests in the tree to work with the new
restriction. That shouldn't have been necessary because we've already seemingly
adjusted all of them to work with the configuration option, but some tests hid
well. We fix that mistake now.
After that, we introduce the new restriction. What's more, when starting Scylla,
we verify that there is no materialized view that would violate the contract.
If there are some that do, we list them, notify the user, and refuse to start.
High-level implementation strategy:
1. Name the restrictions in form of a function.
2. Adjust existing tests.
3. Restrict materialized views by both the experimental feature
and the configuration option. Add validation test.
4. Drop the requirement for the experimental feature. Adjust the added test
and add a new one.
5. Update the user documentation.
Fixesscylladb/scylladb#23030
Backport: 2025.4, as we are aiming to support materialized views for tablets from that version.
Closesscylladb/scylladb#25802
* github.com:scylladb/scylladb:
view: Stop requiring experimental feature
db/view: Verify valid configuration for tablet-based views
db/view: Require rf_rack_valid_keyspaces when creating view
test/cluster/random_failures: Skip creating secondary indexes
test/cluster/mv: Mark test_mv_rf_change as skipped
test/cluster: Adjust MV tests to RF-rack-validity
test/boost/schema_loader_test.cc: Explicitly enable rf_rack_valid_keyspaces
db/view: Name requirement for views with tablets
The querier object is a confusing one. Based on its name it should be in the query/ module and it is already in the query namespace. The query namespace is used for symbols which span the coordinator and replica, or that are mostly coordinator side. The querier is mainly in this namespace due to its similar name and because at the time it was introduced, namespace replica didn't exist yet. But this is a mistake which confuses people.
The querier is actually a completely replica-side logic, implementing the caching of the readers on the replica. Move it to the replica module and namespace to make this more clear.
Code cleanup, no backport.
Closesscylladb/scylladb#26280
* github.com:scylladb/scylladb:
replica: move querier code to replica namespace
root,replica: mv querier to replica/
In preparation for changing their structure.
1) std::map<sstring, sstring> -> replication_strategy_config_options
Parsed options. Values will become std::variant<sstring, rack_list>
2) std::map<sstring, sstring> -> property_definitions::map_type
Flattened map of options, as stored system tables.
We modify the requirements for using materialized views in tablet-based
keyspaces. Before, it was necessary to enable the configuration option
`rf_rack_valid_keyspaces`, having the cluster feature `VIEWS_WITH_TABLETS`
enabled, and using the experimental feature `views-with-tablets`.
We drop the last requirement.
We adjust code to that change and provide a new validation test.
We also update the user documentation to reflect the changes.
Fixesscylladb/scylladb#23030
Creating a materialized view or a secondary index in a tablet-based
keyspace requires that the user enabled two options:
* experimental feature `views-with-tablets`,
* configuration option `rf_rack_vaid_keyspaces`.
Because the latter has only become a necessity recently (in this series),
it's possible that there are already existing materialized views that
violate it.
We add a new check at start-up that iterates over existing views and
makes sure that that is not the case. Otherwise, Scylla notifies the user
of the problem.
`SELECT` commands with SERIAL consistency level are historically allowed for vnode-based views, even though they don't provide linearizability guarantees and in general don't make much sense. In this PR we prohibit LWTs for tablet-based views, but preserve old behavior for vnode-based views for compatibility. Similar logic is applied to CDC log tables.
We also add a general check that disallows colocating a table with another colocated table, since this is not needed for now.
Fixes https://github.com/scylladb/scylladb/issues/26258
backports: not needed (a new feature)
Closesscylladb/scylladb#26284
* github.com:scylladb/scylladb:
cql_test_env.cc: log exception when callback throws
lwt: prohibit for tablet-based views and cdc logs
tablets: disallow chains of colocated tables
database: get_base_table_for_tablet_colocation: extract table_id_by_name lambda
The query namespace is used for symbols which span the coordinator and
replica, or that are mostly coordinator side. The querier is mainly in
this namespace due to its similar name, but this is a mistake which
confuses people. Now that the code was moved to replica/, also fix the
namespace to be namespace replica.
In upcoming commits we’ll add a test to ensure that a table cannot be
colocated with another table that is itself already colocated.
This must also hold in the case where both colocated tables are
created simultaneously in a single migration_manager announcement.
We use Paxos tables as an example of colocated tables in this test.
To support this, get_base_table_for_tablet_colocation needs to look
for the base table among the batch of tables being created.
It belongs there, it is a completely replica-side thing. Also take the
opportunity to rename it to multishard_query.{hh,cc}, it is not just
mutation anymore (data query is also implemented).
Some files in compaction/ have using namespace {compaction,sstables}
clauses, some even in headers. This is considered bad practice and
muddies the namespace use. Remove them.
During an ALTER KEYSPACE statement execution where a table with a view
is present, we need to perform tablet migrations for both tables.
These migrations are not synchronized, so at some point the base may
have a different number of non-pending replicas than the view. Because
of that, we can't pair them correctly. If there is more non-pending
base replicas than view replicas, we don't need to do anything because
the view replica that didn't finish migrating is a pending replica
and will get view updates from all base replicas. But if there is more
non-pending view replicas than base replicas, we may currently lose
view updates to the new view replica.
This patch adds a workaround for this scenario. If after one migration
we have too more non-pending view replicas than base replicas, we add
it to the pending replica list so that it gets an update anyway.
This patch will also take effect if the base and view replica counts
differ due to some other bug. To track that, a new metric is added
to count such occurrences.
This patch also includes a test for this exact scenario, which is enforced by an injection.
Fixes https://github.com/scylladb/scylladb/issues/21492
Our sstable format selection logic is weird, and hard to follow.
If I'm not misunderstanding, the pieces are:
1. There's the `sstable_format` config entry, which currently
doesn't do anything, but in the past it used to disable
cluster features for versions newer than the specified one.
2. There are deprecated and unused config entries for individual
versions (`enable_sstables_mc_format`, `enable_sstables_md_format`,
etc).
3. There is a cluster feature for each version:
ME_SSTABLE_FORMAT, MD_SSTABLE_FORMAT, etc.
(Currently all sstable version features have been grandfathered,
and aren't checked by the code anymore).
4. There's an entry in `system.scylla_local` which contains the
latest enabled sstable version. (Why? Isn't this directly derived
from cluster features anyway)?
5. There's `sstable_manager::_format` which contains the
sstable version to be used for new writes.
This field is updated by `sstables_format_selector`
based on cluster features and the `system.scylla_local` entry.
I don't see why those pieces are needed. Version selection has the
following constraints:
1. New sstables must be written with a format that supports existing
data. For example, range tombstones with an infinite bound are only
supported by sstables since version "mc". So if a range tombstone
with an infinite bound exists somewhere in the dataset,
the format chosen for new sstables has to be at least as new as "mc".
2. A new format might only be used after a corresponding cluster feature
is enabled. (Otherwise new sstables might become unreadable if they
are sent to another node, or if a node is downgraded).
3. The user should have a way to inhibit format ugprades if he wishes.
So far, constraint (1) has been fulfilled by never using formats older
than the newest format ever enabled on the node. (With an exception
for resharding and reshaping system tables).
Constraint (2) has been fulfilled by calling `sstable_manager::set_format`
only after the corresponsing cluster feature is enabled.
Constraint (3) has been fulfilled by the ability to inhibit cluster
features by setting `sstable_format` by some fixed value.
The main thing I don't like about this whole setup is that it doesn't
let me downgrade the preferred sstable format. After a format is
enabled, there is no way to go back to writing the old format again.
That is no good -- after I make some performance-sensitive changes
in a new format, it might turn out to be a pessimization for the
particular workload, and I want to be able to go back.
This patch aims to give a way to downgrade formats without violating
the constraints. What it does is:
1. The entry in `system.scylla_local` becomes obsolete.
After the patch we no longer update or read it.
As far as I understand, the purpose of this entry is to prevent
unwanted format downgrades (which is something cluster features
are designed for) and it's updated if and only if relevant
cluster features are updated. So there's no reason to have it,
we can just directly use cluster features.
2. `sstable_format_selector` gets deleted.
Without the `system.scylla_local` around, it's just a glorified
feature listener.
3. The format selection logic is moved into `sstable_manager`.
It already sees the `db::config` and the `gms::feature_service`.
For the foreseeable future, the knowledge of enabled cluster features
and current config should be enough information to pick the right formats.
4. The `sstable_format` entry in `db::config` is no longer intended to
inhibit cluster features. Instead, it is intended to select the
format for new sstables, and it becomes live-updatable.
5. Instead of writing new sstables with "highest supported" format,
(which used to be set by `sstables_format_selector`) we write
them with the "preferred" format, which is determined by
`sstable_manager` based on the combination of enabled features
and the current value of `sstable_format`.
Closesscylladb/scylladb#26092
[avi: Pavel found the reason for the scylla_local entry -
it predates stable storage for cluster features]
The latter is recommended in seastar, and the former was left as
compatibility alias. Latest seastar explicitly marks it as deprecated so
once the submodule is updated, compilation logs will explode.
Most of the patch is generated with
for f in $(git grep -l '\<distributed<[A-Za-z0-9:_]*>') ; do sed -e 's/\<distributed<\([A-Za-z0-9:_]*\)>/sharded<\1>/g' -i $f; done
for f in $(git grep -l distributed.hh); do sed -e 's/distributed.hh/sharded.hh/' -i $f ; done
and a small manual change in test/perf/perf.hh
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#26136
When creating a tablet map for a CDC table, make it be co-located with
its base table.
We modify db::get_base_table_for_tablet_colocation to return the base
table id of a CDC table, handling both cases that the base table is a
new table that's created in the same operation, or is an existing table
in the db. This function is used by the tablet allocator to decide
whether to create a co-located tablet map or allocate new tablets.
Fixes#25683
Once a table drop is complete, there should be no reason to retain
truncation records for it, as any replay should skip mutations
anyway (no CF), and iff we somehow resurrect a dropped table,
this replay-resurrected data is the least problem anyway.
Adds a prune phase to the startup drop_truncation_rp_records run,
which ignores updating, and instead deletes records for non-existant
tables (which should patch any existing servers with lingering data
as well).
Also does an explicit delete of records on actual table DROP, to
ensure we don't grow this table more than needed even in long
uptime nodes.
Small unit test included.
Closesscylladb/scylladb#25699
When database operates in the critical disk utilization mode, all
mutation writes including inserts, updates, deletes, counter updates,
hints, read+repair, lwt writes) to user tables and other associated
with them tables like views, CDC log, audit are rejected, with a clear
error exception returned.
The mode is meant to be used with the disk space monitor in order
to prevent any user writes when node's disk utilization is too high.
When the configuration option `rf_rack_valid_keyspaces` is enabled and there
is an RF-rack-invalid keyspace, starting a node fails. However, when the
configuration option is disabled, but there still is a keyspace that violates
the condition, we'd like Scylla to print a warning informing the user about
the fact. That's what happens in this commit.
We provide a validation test.
The central idea of incremental repair is to allow repair participants
to select and repair only a portion of the dataset to speed up the
repair process. All repair participants must utilize an identical
selection method to repair and synchronize the same selected dataset.
There are two primary selection methods: time-based and file-based. The
time-based method selects data within a specified time frame. It is
versatile but it is less efficient because it requires reading all of
the dataset and omitting data beyond the time frame. The file-based
method selects data from unrepaired SSTables and is more efficient
because it allows the entire SSTable to be omitted. This document patch
implements the file-based selection method.
Incremental repair will only be supported for tablet tables; it will not
be supported for vnode tables. On one hand, the legacy vnode is less
important to support. On the other hand, the incremental repair for
vnode is much harder to implement. With vnodes, a SSTalbe could contain
data for multiple vnode ranges. When a given vnode range is repaired,
only a portion of the SSTable is repaired. This complicates the
manipulation of SSTables significantly during both repair and
compaction. With tablets, an entire tablet is repaired so that a
sstable is either fully repaired or not repaired which is a huge
simplification.
This patch uses the repaired_at from sstables::statistics component to
mark a sstable as repaired. It uses a virtual clock as the repair
timestamp, i.e., using a monotonically increasing number for the
repaired_at field of a SSTable and sstables_repaired_at column in
system.tablets table. Notice that when a sstable is not repaired, the
repaired_at field will be set to the default value 0 by default. The
being_repaired in memory field of a SSTable is used to explicitly mark
that a SSTable is being selected. The following variables are used for
incremental repair:
The repaired_at on disk field of a SSTable is used.
- A 64-bit number increases sequentially
The sstables_repaired_at is added to the system.tablets table.
- repaired_at <= sstables_repaired_at means the sstable is repaired
The being_repaired in memory field of a SSTable is added.
- A repair UUID tells which sstable has participated in the repair
Initial test results:
1) Medium dataset results
Node amount: 3
Instance type: i4i.2xlarge
Disk usage per node: ~500GB
Cluster pre-populated with ~500GB of data before starting repairs job.
Results for Repair Timings:
The regular repair run took 210 mins.
Incremental repair 1st run took 183 mins, 2nd and 3rd runs took around 48s
The speedup is: 183 mins / 48s = 228X
2) Small dataset results
Node amount: 3
Instance type: i4i.2xlarge
Disk usage per node: ~167GB
Cluster pre-populated with ~167GB of data before starting the repairs job.
Regular repair 1st run took 110s, 2nd and 3rd runs took 110s.
Incremental repair 1st run took 110 seconds, 2nd and 3rd run took 1.5 seconds.
The speedup is: 110s / 1.5s = 73X
3) Large dataset results
Node amount: 6
Instance type: i4i.2xlarge, 3 racks
50% of base load, 50% read/write
Dataset == Sum of data on each node
Dataset Non-incremental repair (minutes)
1.3 TiB 31:07
3.5 TiB 25:10
5.0 TiB 19:03
6.3 TiB 31:42
Dataset Incremental repair (minutes)
1.3 TiB 24:32
3.0 TiB 13:06
4.0 TiB 5:23
4.8 TiB 7:14
5.6 TiB 3:58
6.3 TiB 7:33
7.0 TiB 6:55
Fixes#22472Closesscylladb/scylladb#24291
* github.com:scylladb/scylladb:
replica: Introduce get_compaction_reenablers_and_lock_holders_for_repair
compaction: Move compaction_reenabler to compaction_reenabler.hh
topology_coordinator: Make rpc::remote_verb_error to warning level
repair: Add metrics for sstable bytes read and skipped from sstables
test.py: Disable incremental for test_tombstone_gc_for_streaming_and_repair
test.py: Add tests for tablet incremental repair
repair: Add tablet incremental repair support
compaction: Add tablet incremental repair support
feature_service: Add TABLET_INCREMENTAL_REPAIR feature
tablet_allocator: Add tablet_force_tablet_count_increase and decrease
repair: Add incremental helpers
sstable: Add being_repaired to sstable
sstables: Add set_repaired_at to metadata_collector
mutation_compactor: Introduce add operator to compaction_stats
tablet: Add sstables_repaired_at to system.tablets table
test: Fix drain api in task_manager_client.py
Instead of storing it partially in tombstone_gc and partially in an
external map. Move all external parts into the new
shared_tombstone_gc_state. This new class is responsible for
keeping and updating the repair history. tombstone_gc_state just keeps
const pointers to the shared state as before and is only responsible for
querying the tombstone gc before times.
This separation makes the code easier to follow and also enables further
patching of tombstone_gc_state.
Since table_state is a view to a compaction group, it makes sense
to rename it as so.
With upcoming incremental repair, each replica::compaction_group
will be actually two compaction groups, so there will be two
views for each replica::compaction_group.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
to *_static_effective_replication_map, in preparation
for separating local_effective_replication_map from
vnode_effective_replication_map.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
