in 7952200c, we changed the `selected_format` from `mc` to `me`,
but to be backward compatible the cluster starts with "md", so
when the nodes in cluster agree on the "ME_SSTABLE_FORMAT" feature,
the format selector believes that the node is already using "ME",
which is specified by `_selected_format`. even it is actually still
using "md", which is specified by `sstable_manager::_format`, as
changed by 54d49c04. as explained above, it was specified to "md"
in hope to be backward compatible when upgrading from an existign
installation which might be still using "md". but after a second
thought, since we are able to read sstables persisted with older
formats, this concern is not valid.
in other words, 7952200c introduced a regression which changed the
"default" sstable format from `me` to `md`.
to address this, we just change `sstable_manager::_format` to "me",
so that all sstables are created using "me" format.
a test is added accordingly.
Fixes#18995
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
(cherry picked from commit 5a0d30f345)
Closesscylladb/scylladb#19422
Normally, the space overhead for TWCS is 1/N, where is number of windows. But during off-strategy, the overhead is 100% because input sstables cannot be released earlier.
Reshaping a TWCS table that takes ~50% of available space can result in system running out of space.
That's fixed by restricting every TWCS off-strategy job to 10% of free space in disk. Tables that aren't big will not be penalized with increased write amplification, as all input (disjoint) sstables can still be compacted in a single round.
Fixes#16514.
(cherry picked from commit b8bd4c51c2)
(cherry picked from commit 51c7ee889e)
(cherry picked from commit 0ce8ee03f1)
(cherry picked from commit ace4e5111e)
Refs #18137Closesscylladb/scylladb#19404
* github.com:scylladb/scylladb:
compaction: Reduce twcs off-strategy space overhead to 10% of free space
compaction: wire storage free space into reshape procedure
sstables: Allow to get free space from underlying storage
replica: don't expose compaction_group to reshape task
This test in topology_experimental_raft/test_alternator.py wants to
check that during Alternator TTL's expiration scans, ALL of the CPU was
used in the "streaming" scheduling group and not in the "statement"
scheduling group. But to allow for some fluke requests (e.g., from the
driver), the test actually allows work in the statement group to be
up to 1% of the work.
Unfortunately, in one test run - a very slow debug+aarch64 run - we
saw the work on the statement group reach 1.4%, failing the test.
I don't know exactly where this work comes from, perhaps the driver,
but before this bug was fixed we saw more than 58% of the work in the
wrong scheduling group, so neither 1% or 1.4% is a sign that the bug
came back. In fact, let's just change the threshold in the test to 10%,
which is also much lower than the pre-fix value of 58%, so is still a
valid regression test.
Fixes#19307
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
(cherry picked from commit 9fc70a28ca)
Closesscylladb/scylladb#19333
TWCS off-strategy suffers with 100% space overhead, so a big TWCS table
can cause scylla to run out of disk space during node ops.
To not penalize TWCS tables, that take a small percentage of disk,
with increased write ampl, TWCS off-strategy will be restricted to
10% of free disk space. Then small tables can still compact all
disjoint sstables in a single round.
Fixes#16514.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
(cherry picked from commit ace4e5111e)
After this, TWCS reshape procedure can be changed to limit job
to 10% of available space.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
(cherry picked from commit 0ce8ee03f1)
That will be used in turn to restrict reshape to 10% of available space
in underlying storage.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
(cherry picked from commit 51c7ee889e)
Fixesscylladb/scylla-pkg#3845
Don't overwrite (or rather change) AWS credentials variables if already set in
enclosing environment. Ensures EAR tests for AWS KMS can run properly in CI.
v2:
* Allow environment variables in reading obj storage config - allows CI to
use real credentials in env without risking putting them info less seure
files
* Don't write credentials info from miniserver into config, instead use said
environment vars to propagate creds.
v3:
* Fix python launch scripts to not clear environment, thus retaining above aws envs.
(cherry picked from commit 5056a98289)
Closesscylladb/scylladb#19330
utils/chunked_vector::reserve_partial: fix usage in callers
The method reserve_partial(), when used as documented, quits before the
intended capacity can be reserved fully. This can lead to overallocation
of memory in the last chunk when data is inserted to the chunked vector.
The method itself doesn't have any bug but the way it is being used by
the callers needs to be updated to get the desired behaviour.
Instead of calling it repeatedly with the value returned from the
previous call until it returns zero, it should be repeatedly called with
the intended size until the vector's capacity reaches that size.
This PR updates the method comment and all the callers to use the
right way.
Fixes#19254
(cherry picked from commit 64768b58e5)
(cherry picked from commit 29f036a777)
(cherry picked from commit 0a22759c2a)
(cherry picked from commit d4f8b91bd6)
(cherry picked from commit 310c5da4bb)
(cherry picked from commit 83190fa075)
(cherry picked from commit c49f6391ab)
Refs #19279Closesscylladb/scylladb#19310
* github.com:scylladb/scylladb:
utils/large_bitset: remove unused includes identified by clangd
utils/large_bitset: use thread::maybe_yield()
test/boost/chunked_managed_vector_test: fix testcase tests_reserve_partial
utils/lsa/chunked_managed_vector: fix reserve_partial()
utils/chunked_vector: return void from reserve_partial and make_room
test/boost/chunked_vector_test: fix testcase tests_reserve_partial
utils/chunked_vector::reserve_partial: fix usage in callers
Update the maximum size tested by the testcase. The test always created
only one chunk as the maximum size tested by it (1 << 12 = 4KB) was less
than the default max chunk size (12.8 KB). So, use twice the
max_chunk_capacity as the test size distribution upper limit to verify
that partial_reserve can reserve multiple chunks.
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
(cherry picked from commit 310c5da4bb)
Fix the method comment and return types of chunked_managed_vector's
reserve_partial() similar to chunked_vector's reserve_partial() as it
has the same issues mentioned in #19254. Also update the usage in the
chunked_managed_vector_test.
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
(cherry picked from commit d4f8b91bd6)
Fix the usage of reserve_partial in the testcase. Also update the
maximum chunk size used by the testcase. The test always created only
one chunk as the maximum size tested by it (1 << 12 = 4KB) was less
than the default max chunk size (128 KB). So, use smaller chunk size,
512 bytes, to verify that partial_reserve can reserve multiple chunks.
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
(cherry picked from commit 29f036a777)
in e7d4e080, we reenabled the background writes in this test, but
when running with tablets enabled, background writes are still
disabled because of #17025, which was fixed last week. so we can
enable background writes with tablets.
in this change,
* background writes are enabled with tablets.
* increase the number of nodes by 1 so that we have enough nodes
to fulfill the needs of tablets, which enforces that the number
of replicas should always satisfy RF.
* pass rf to `start_writes()` explicitly, so we have less
magic numbers in the test, and make the data dependencies
more obvious.
Fixes#17589
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
(cherry picked from commit 77f0259a63)
Closesscylladb/scylladb#19184
This config item is propagated to the table object via table::config. Although the field in `table::config`, used to propagate the value, was `utils::updateable_value<T>`, it was assigned a constant and so the live-update chain was broken.
This series fixes this and adds a test which fails before the patch and passes after. The test needed new test infrastructure, around the failure injection api, namely the ability to exfiltrate the value of internal variable. This infrastructure is also added in this series.
Fixes: https://github.com/scylladb/scylladb/issues/18674
- [x] This patch has to be backported because it fixes broken functionality
(cherry picked from commit dbccb61636)
(cherry picked from commit 4590026b38)
(cherry picked from commit feea609e37)
(cherry picked from commit 0c61b1822c)
(cherry picked from commit 8ef4fbdb87)
Refs #18705Closesscylladb/scylladb#19240
* github.com:scylladb/scylladb:
test/topology_custom: add test for enable_compacting_data_for_streaming_and_repair live-update
test/pylib: rest_client: add get_injection()
api/error_injection: add getter for error_injection
utils/error_injection: add set_parameter()
replica/database: fix live-update enable_compacting_data_for_streaming_and_repair
before this change, when performing memtable_test, we expect that
the memtables of ks.cf is the only memtables being flushed. and
we inject 4 failures in the code path of flush, and wait until 4
of them are triggered. but in the background, `dirty_memory_manager`
performs flush on all tables when necessary. so, the total number of
failures is not necessary the total number of failures triggered
when flushing ks.cf, some of them could be triggered when flushing
system tables. that's why we have sporadict test failures from
this test. as we might check `t.min_memtable_timestamp()` too soon.
after this change, we increase `unspooled_dirty_soft_limit` setting,
in order to disable `dirty_memory_manager`, so that the only flush
is performed by the test.
Fixes#19034
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
(cherry picked from commit 223fba3243)
before this change, we verify the behavior of design under test using
`BOOST_ASSERT()`, which is a wrapper around `assert()`, so if a test
fails, the test just aborts. this is not very helpful for postmortem
debugging.
after this change, we use `BOOST_REQUIRE` macro for verifying the
behavior, so that Boost.Test prints out the condition if it does not
hold when we test it.
Refs #19034
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
(cherry picked from commit 2df4e9cfc2)
Most of the time this test spends waiting for a node to die. Helps 3x times
Was
real 9m21,950s
user 1m11,439s
sys 1m26,022s
Now
real 3m37,780s
user 0m58,439s
sys 1m13,698s
refs: #17764
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
(cherry picked from commit a4e8f9340a)
Closesscylladb/scylladb#19233
The API already promises this, the comment on effective_replication_map says:
"Excludes replicas which are in the left state".
Tablet replicas on the replaced node are rebuilt after the node
already left. We may no longer have the IP mapping for the left node
so we should not include that node in the replica set. Otherwise,
storage_proxy may try to use the empty IP and fail:
storage_proxy - No mapping for :: in the passed effective replication map
It's fine to not include it, because storage proxy uses keyspace RF
and not replica list size to determine quorum. The node is not coming
up, so noone should need to contact it.
Users which need replica list stability should use the host_id-based API.
Fixes#18843
(cherry picked from commit 3e1ba4c859)
(cherry picked from commit 0d596a425c)
Refs #18955Closesscylladb/scylladb#19143
* github.com:scylladb/scylladb:
tablets: Filter-out left nodes in get_natural_endpoints()
test: pylib: Extract start_writes() load generator utility
Consider the following:
1) table A has N tablets and views
2) migration starts for a tablet of A from node 1 to 2.
3) migration is at write_both_read_old stage
4) coordinator will push writes to both nodes (pending and leaving)
5) A has view, so writes to it will also result in reads (table::push_view_replica_updates())
6) tablet's update_effective_replication_map() is not refreshing tablet sstable set (for new tablet migrating in)
7) so read on step 5 is not being able to find sstable set for tablet migrating in
Causes the following error:
"tablets - SSTable set wasn't found for tablet 21 of table mview.users"
which means loss of write on pending replica.
The fix will refresh the table's sstable set (tablet_sstable_set) and cache's snapshot.
It's not a problem to refresh the cache snapshot as long as the logical
state of the data hasn't changed, which is true when allocating new
tablet replicas. That's also done in the context of compactions for example.
Fixes#19052.
Fixes#19033.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
(cherry picked from commit 7b41630299)
Closesscylladb/scylladb#19229
The API already promises this, the comment on effective_replication_map says:
"Excludes replicas which are in the left state".
Tablet replicas on the replaced node are rebuilt after the node
already left. We may no longer have the IP mapping for the left node
so we should not include that node in the replica set. Otherwise,
storage_proxy may try to use the empty IP and fail:
storage_proxy - No mapping for :: in the passed effective replication map
It's fine to not include it, because storage proxy uses keyspace RF
and not replica list size to determine quorum. The node is not coming
up, so noone should need to contact it.
Users which need replica list stability should use the host_id-based API.
Fixes#18843
(cherry picked from commit 0d596a425c)
The Alternator test test_metrics.py::test_item_latency confirms that
for several operation types (PutItem, GetItem, DeleteItem, UpdateItem)
we did not forget to measure their latencies.
The test checked that a latency was updated by checking that two metrics
increases:
scylla_alternator_op_latency_count
scylla_alternator_op_latency_sum
However, it turns out that the "sum" is only an approximate sum of all
latencies, and when the total sum grows large it sometimes does *not*
increase when a short latency is added to the statistics. When this
happens, this test fails on the assertion that the "sum" increases after
an operation. We saw this happening sometimes in CI runs.
The simple fix is to stop checking _sum at all, and only verify that
the _count increases - this is really an integer counter that
unconditionally increases when a latency is added to the histogram.
Don't worry that the strength of this test is reduced - this test was
never meant to check the accuracy or correctness of the histograms -
we should have different (and better) tests for that, unrelated to
Alternator. The purpose of *this* test is only to verify that for some
specific operation like PutItem, Alternator didn't forget to measure its
latency and update the histogram. We want to avoid a bug like we had
in counters in the past (#9406).
Fixes#18847.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
(cherry picked from commit 13cf6c543d)
Closesscylladb/scylladb#19193
The check query may be executed on a node which doesn't yet see that
the downed server is down, as it is not shut down gracefully. The
query coordinator can choose the down node as a CL=1 replica for read
and time out.
To fix, wait for all nodes to notice the node is down before executing
the checking query.
Fixes#17938
(cherry picked from commit c8f71f4825)
Closesscylladb/scylladb#19199
Alternator has a custom TTL implementation. This is based on a loop, which scans existing rows in the table, then decides whether each row have reached its end-of-life and deletes it if it did. This work is done in the background, and therefore it uses the maintenance (streaming) scheduling group. However, it was observed that part of this work leaks into the statement scheduling group, competing with user workloads, negatively affecting its latencies. This was found to be causes by the reads and writes done on behalf of the alternator TTL, which looses its maintenance scheduling group when these have to go to a remote node. This is because the messaging service was not configured to recognize the streaming scheduling group, when statement verbs like read or writes are invoked. The messaging service currently recognizes two statement "tenants": the user tenant (statement scheduling group) and system (default scheduling group), as we used to have only user-initiated operations and sytsem (internal) ones. With alternator TTL, there is now a need to distinguish between two kinds of system operation: foreground and background ones. The former should use the system tenant while the latter will use the new maintenance tenant (streaming scheduling group).
This series adds a streaming tenant to the messaging service configuration and it adds a test which confirms that with this change, alternator TTL is entirely contained in the maintenance scheduling group.
Fixes: #18719
- [x] Scans executed on behalf of alternator TTL are running in the statement group, disturbing user-workloads, this PR has to be backported to fix this.
(cherry picked from commit 5d3f7c13f9)
(cherry picked from commit 1fe8f22d89)
Refs #18729Closesscylladb/scylladb#19196
* github.com:scylladb/scylladb:
alternator, scheduler: test reproducing RPC scheduling group bug
main: add maintenance tenant to messaging_service's scheduling config
Currently they both run in streaming group and it may become busy during
repair/mv building and affect group0 functionality. Move it to the
gossiper group where it should have more time to run.
Fixes#18863
(cherry picked from commit a74fbab99a)
Closesscylladb/scylladb#19175
This patch adds a test for issue #18719: Although the Alternator TTL
work is supposedly done in the "streaming" scheduling group, it turned
out we had a bug where work sent on behalf of that code to other nodes
failed to inherit the correct scheduling group, and was done in the
normal ("statement") group.
Because this problem only happens when more than one node is involved,
the test is in the multi-node test framework test/topology_experimental_raft.
The test uses the Alternator API. We already had in that framework a
test using the Alternator API (a test for alternator+tablets), so in
this patch we move the common Alternator utility functions to a common
file, test_alternator.py, where I also put the new test.
The test is based on metrics: We write expiring data, wait for it to expire,
and then check the metrics on how much CPU work was done in the wrong
scheduling group ("statement"). Before #18719 was fixed, a lot of work
was done there (more than half of the work done in the right group).
After the issue was fixed in the previous patch, the work on the wrong
scheduling group went down to zero.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
(cherry picked from commit 1fe8f22d89)
We want to exclude repair with tablet migrations to avoid races
between repair reads and writes with replica movement. Repair is not
prepared to handle topology transitions in the middle.
One reason why it's not safe is that repair may successfully write to
a leaving replica post streaming phase and consider all replicas to be
repaired, but in fact they are not, the new replica would not be
repaired.
Other kinds of races could result in repair failures. If repair writes
to a leaving replica which was already cleaned up, such writes will
fail, causing repair to fail.
Excluding works by keeping effective_replication_map_ptr in a version
which doesn't have table's tablets in transitions. That prevents later
transitions from starting because topology coordinator's barrier will
wait for that erm before moving to a stage later than
allow_write_both_read_old, so before any requests start using the new
topology. Also, if transitions are already running, repair waits for
them to finish.
A blocked tablet migration (e.g. due to down node) will block repair,
whereas before it would fail. Once admin resolves the cause of blocked migration,
repair will continue.
Fixes#17658.
Fixes#18561.
(cherry picked from commit 6c64cf33df)
(cherry picked from commit 1513d6f0b0)
(cherry picked from commit 476c076a21)
(cherry picked from commit c45ce41330)
(cherry picked from commit e97acf4e30)
(cherry picked from commit 98323be296)
(cherry picked from commit 5ca54a6e88)
Refs #18641Closesscylladb/scylladb#19144
* github.com:scylladb/scylladb:
test: pylib: Do not block async reactor while removing directories
repair: Exclude tablet migrations with tablet repair
repair_service: Propagate topology_state_machine to repair_service
main, storage_service: Move topology_state_machine outside storage_service
storage_srvice, toplogy: Extract topology_state_machine::await_quiesced()
tablet_scheduler: Make disabling of balancing interrupt shuffle mode
tablet_scheduler: Log whether balancing is considered as enabled
This fixes a problem where suite cleanup schedules lots of uninstall()
tasks for servers started in the suite, which schedules lots of tasks,
which synchronously call rmtree(). These take over a minute to finish,
which blocks other tasks for tests which are still executing.
In particular, this was observed to case
ManagerClient.server_stop_gracefully() to time-out. It has a timeout
of 60 seconds. The server was stopped quickly, but the RESTful API
response was not processed in time and the call timed out when it got
the async reactor.
(cherry picked from commit 5ca54a6e88)
We want to exclude repair with tablet migrations to avoid races
between repair reads and writes with replica movement. Repair is not
prepared to handle topology transitions in the middle.
One reason why it's not safe is that repair may successfully write to
a leaving replica post streaming phase and consider all replicas to be
repaired, but in fact they are not, the new replica would not be
repaired.
Other kinds of races could result in repair failures. If repair writes
to a leaving replica which was already cleaned up, such writes will
fail, causing repair to fail.
Excluding works by keeping effective_replication_map_ptr in a version
which doesn't have table's tablets in transitions. That prevents later
transitions from starting because topology coordinator's barrier will
wait for that erm before moving to a stage later than
allow_write_both_read_old, so before any requets start using the new
topology. Also, if transitions are already running, repair waits for
them to finish.
Fixes#17658.
Fixes#18561.
(cherry picked from commit 98323be296)
If a node restart just before it stores bootstrapping node's IP it will
not have ID to IP mapping for bootstrapping node which may cause failure
on a write path. Detect this and fail bootstrapping if it happens.
(cherry picked from commit 1faef47952)
(cherry picked from commit 27445f5291)
(cherry picked from commit 6853b02c00)
(cherry picked from commit f91db0c1e4)
Refs #18927Closesscylladb/scylladb#19118
* github.com:scylladb/scylladb:
raft topology: fix indentation after previous commit
raft topology: do not add bootstrapping node without IP as pending
test: add test of bootstrap where the coordinator crashes just before storing IP mapping
schema_tables: remove unused code
Fetching only the first page is not the intuitive behavior expected by users.
This causes flakiness in some tests which generate variable amount of
keys depending on execution speed and verify later that all keys were
written using a single SELECT statement. When the amount of keys
becomes larger than page size, the test fails.
Fixes#18774
(cherry picked from commit 2c3f7c996f)
Closesscylladb/scylladb#19130
Task manager's tasks stay in memory after they are finished.
Moreover, even if a child task is unregistered from task manager,
it is still alive since its parent keeps a foreign pointer to it. Also,
when a task has finished successfully there is no point in keeping
all of its descendants in memory.
The patch introduces folding of task manager's tasks. Whenever
a task which has a parent is finished it is unregistered from task
manager and foreign_ptr to it (kept in its parent) is replaced
with its status. Children's statuses of the task are dropped unless
they or one of their descendants failed. So for each operation we
keep a tree of tasks which contains:
- a root task and its direct children (status if they are finished, a task
otherwise);
- running tasks and their direct children (same as above);
- a statuses path from root to failed tasks.
/task_manager/wait_task/ does not unregister tasks anymore.
Refs: https://github.com/scylladb/scylladb/issues/16694.
- [ ] ** Backport reason (please explain below if this patch should be backported or not) **
Requires backport to 6.0 as task number exploded with tablets.
(cherry picked from commit 6add9edf8a)
(cherry picked from commit 319e799089)
(cherry picked from commit e6c50ad2d0)
(cherry picked from commit a82a2f0624)
(cherry picked from commit c1b2b8cb2c)
(cherry picked from commit 30f97ea133)
(cherry picked from commit fc0796f684)
(cherry picked from commit d7e80a6520)
(cherry picked from commit beef77a778)
Refs https://github.com/scylladb/scylladb/pull/18735Closesscylladb/scylladb#19104
* github.com:scylladb/scylladb:
docs: describe task folding
test: rest_api: add test for task tree structure
test: rest_api: modify new_test_module
tasks: test: modify test_task methods
api: task_manager: do not unregister task in /task_manager/wait_task/
tasks: unregister tasks with parents when they are finished
tasks: fold finished tasks info their parents
tasks: make task_manager::task::impl::finish_failed noexcept
tasks: change _children type
On the next boot there is no host ID to IP mapping which causes node to
crash again with "No mapping for :: in the passed effective replication map"
assertion.
(cherry picked from commit 27445f5291)
The values of `tablets_enabled` were nonempty strings, so they
always evaluated to `True` in the if statement responsible for
enabling writing workers only if tablets are disabled. Hence, the
writing workers were always disabled.
The original commit, ea4717da65,
contains one more change, which is not needed (and conflicting)
in 6.0 because scylladb/scylladb#18898 has been backported first.
Closesscylladb/scylladb#19111
Retrieval of tablet stats must be serialized with mutation to token metadata, as the former requires tablet id stability.
If tablet split is finalized while retrieving stats, the saved erm, used by all shards, can have a lower tablet count than the one in a particular shard, causing an abort as tablet map requires that any id feeded into it is lower than its current tablet count.
Fixes https://github.com/scylladb/scylladb/issues/18085.
(cherry picked from commit abcc68dbe7)
(cherry picked from commit 551bf9dd58)
(cherry picked from commit e7246751b6)
Refs https://github.com/scylladb/scylladb/pull/18287Closesscylladb/scylladb#19095
* github.com:scylladb/scylladb:
topology_experimental_raft/test_tablets: restore usage of check_with_down
test: Fix flakiness in topology_experimental_raft/test_tablets
service: Use tablet read selector to determine which replica to account table stats
storage_service: Fix race between tablet split and stats retrieval
In order to avoid per-table tablet load imbalance balance from forming
in the cluster after adding nodes, the load balancer now picks the
candidate tablet at random. This should keep the per-table
distribution on the target node similar to the distribution on the
source nodes.
Currently, candidate selection picks the first tablet in the
unordered_set, so the distribution depends on hashing in the unordered
set. Due to the way hash is calculated, table id dominates the hash
and a single table can be chosen more often for migration away. This
can result in imbalance of tablets for any given table after
bootstrapping a new node.
For example, consider the following results of a simulation which
starts with a 6-node cluster and does a sequence of node bootstraps
and decommissions. One table has 4096 tablets and RF=1, and the other
has 256 tablets and RF=2. Before the patch, the smaller table has
node overcommit of 2.34 in the worst topology state, while after the
patch it has overcommit of 1.65. overcommit is calculated as max load
(tablet count per node) dividied by perfect average load (all tablets / nodes):
Run #861, params: {iterations=6, nodes=6, tablets1=4096 (10.7/sh), tablets2=256 (1.3/sh), rf1=1, rf2=2, shards=64}
Overcommit : init : {table1={shard=1.03, node=1.00}, table2={shard=1.51, node=1.01}}
Overcommit : worst: {table1={shard=1.23, node=1.10}, table2={shard=9.85, node=1.65}}
Overcommit (old) : init : {table1={shard=1.03, node=1.00}, table2={shard=1.51, node=1.01}}
Overcommit (old) : worst: {table1={shard=1.31, node=1.12}, table2={shard=64.00, node=2.34}}
The worst state before the patch had the following distribution of tablets for the smaller table:
Load on host ba7f866d...: total=171, min=1, max=7, spread=6, avg=2.67, overcommit=2.62
Load on host 4049ae8d...: total=102, min=0, max=6, spread=6, avg=1.59, overcommit=3.76
Load on host 3b499995...: total=89, min=0, max=4, spread=4, avg=1.39, overcommit=2.88
Load on host ad33bede...: total=63, min=0, max=3, spread=3, avg=0.98, overcommit=3.05
Load on host 0c2e65dc...: total=57, min=0, max=3, spread=3, avg=0.89, overcommit=3.37
Load on host 3f2d32d4...: total=27, min=0, max=2, spread=2, avg=0.42, overcommit=4.74
Load on host 9de9f71b...: total=3, min=0, max=1, spread=1, avg=0.05, overcommit=21.33
One node has as many as 171 tablets of that table and another one has as few as 3.
After the patch, the worst distribution looks like this:
Load on host 94a02049...: total=121, min=1, max=6, spread=5, avg=1.89, overcommit=3.17
Load on host 65ac6145...: total=87, min=0, max=5, spread=5, avg=1.36, overcommit=3.68
Load on host 856a66d1...: total=80, min=0, max=5, spread=5, avg=1.25, overcommit=4.00
Load on host e3ac4a41...: total=77, min=0, max=4, spread=4, avg=1.20, overcommit=3.32
Load on host 81af623f...: total=66, min=0, max=4, spread=4, avg=1.03, overcommit=3.88
Load on host 4a038569...: total=47, min=0, max=2, spread=2, avg=0.73, overcommit=2.72
Load on host c6ab3fe9...: total=34, min=0, max=3, spread=3, avg=0.53, overcommit=5.65
Most-loaded node has 121 tablets and least loaded node has 34 tablets.
It's still not good, a better distribution is possible, but it's an improvement.
Refs #16824
(cherry picked from commit 3be6120e3b)
(cherry picked from commit c9bcb5e400)
(cherry picked from commit 7b1eea794b)
(cherry picked from commit 603abddca9)
Refs #18885Closesscylladb/scylladb#19036
* github.com:scylladb/scylladb:
tablets: load balancer: Use random selection of candidates when moving tablets
test: perf: Add test for tablet load balancer effectiveness
load_sketch: Extract get_shard_minmax()
load_sketch: Allow populating only for a given table
Incremented the components_memory_reclaim_threshold config's default
value to 0.2 as the previous value was too strict and caused unnecessary
eviction in otherwise healthy clusters.
Fixes#18607
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
(cherry picked from commit 3d7d1fa72a)
Closesscylladb/scylladb#19014
... and replace it with boolean enable_tablets option. All the places
in the code are patched to check the latter option instead of the former
feature.
The option is OFF by default, but the default scylla.yaml file sets this
to true, so that newly installed clusters turn tablets ON.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
(cherry picked from commit 83d491af02)
Closesscylladb/scylladb#19012
Separate keyspace which also behaves as system brings
little benefit while creating some compatibility problems
like schema digest mismatch during rollback. So we decided
to move auth tables into system keyspace.
Fixes https://github.com/scylladb/scylladb/issues/18098Closesscylladb/scylladb#18769
(cherry picked from commit 2ab143fb40)
[avi: adjust test/alternator/suite.yaml to reflect new keyspace]
When calculating the base-view mapping while the topology
is changing, we may encounter a situation where the base
table noticed the change in its effective replication map
while the view table hasn't, or vice-versa. This can happen
because the ERM update may be performed during the preemption
between taking the base ERM and view ERM, or, due to f2ff701,
the update may have just been performed partially when we are
taking the ERMs.
Until now, we assumed that the ERMs are synchronized while calling
finding the base-view endpoint mapping, so in particular, we were
using the topology from the base's ERM to check the datacenters of
all endpoints. Now that the ERMs are more likely to not be the same,
we may try to get the datacenter of a view endpoint that doesn't
exist in the base's topology, causing us to crash.
This is fixed in this patch by using the view table's topology for
endpoints coming from the view ERM. The mapping resulting from the
call might now be a temporary mapping between endpoints in different
topologies, but it still maps base and view replicas 1-to-1.
Fixes: #17786Fixes: #18709
(cherry-picked from 519317dc58)
This commit also includes the follow-up patch that removes the
flakiness from the test that is introduced by the commit above.
The flakiness was caused by enabling the
delay_before_get_view_natural_endpoint injection on a node
and not disabling it before the node is shut down. The patch
removes the enabling of the injection on the node in the first
place.
By squashing the commits, we won't introduce a place in the
commit history where a potential bisect could mistakenly fail.
Fixes: https://github.com/scylladb/scylladb/issues/18941
(cherry-picked from 0de3a5f3ff)
Closesscylladb/scylladb#18974
The check is performed by selecting from mutation_fragments(table), but
it's known that this query crashes Scylla when there's no tablet replica
on that node.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
When the test changes RF from 2 to 3, the extra node executes "rebuild"
transition which means that it streams tablets replicas from two other
peers. When doing it, the node receives two sets of sstables with
mutations from the given tablet. The test part that checks if the extra
node received the mutations notices two mutation fragments on the new
replica and errorneously fails by seeing, that RF=3 is not equal to the
number of mutations found, which is 4.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
