these unused includes were identified by clang-include-cleaner. after
auditing these source files, all of the reports have been confirmed.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
In this rather large path we mode to address nodes in storage proxy by
host ids instead of ips. Some subsystems storage proxy calls to are
not yet converted to host ids, so we translate back and forth when we
interact with them.
In this commit, we add two new metrics to storage proxy:
* `received_hints_total`,
* `received_hints_bytes_total`.
Before these changes, we had to rely solely on other
metrics indicating how many hints nodes have written,
rejected, sent, etc. Because hints are subject to
many more or less controllable factors, e.g. a target
node still being a replica for a mutation, it was
very difficult to approximate how many hints a given
node might have received or what part of its load
they were. The newly introduced metrics are supposed
to help reason about those.
There are two metrics to help observe base-write throttling:
* current_throttled_base_writes
* last_mv_flow_control_delay
Both show a snapshot of what is happening right at the time of querying
these metrincs. This doesn't work well when one wants to investigate the
role throttling is playing in occasional write timeouts.s Prometheus
scrapes metrics in multi-second intervals, and the probability of that
instant catching the throttling at play is very small (almost zero).
Add two new metrics:
* throttled_base_writes_total
* mv_flow_control_delay_total
These accumulate all values, allowing graphana to derive the values and
extract information about throttle events that happened in the past
(but not necessarily at the instant of the scrape).
Note that dividing the two values, will yield the average delay for a
throttle, which is also useful.
Closesscylladb/scylladb#18435
Fixes#11017
When doing writes, storage proxy creates types deriving from abstract_write_response_handler.
These are created in the various scheduling groups executing the write inducing code. They
pick up a group-local reference to the various metrics used by SP. Normally all code
using (and esp. modifying) these metrics are executed in the same scheduling group.
However, if gossip sees a node go down, it will notify listeners, which eventually
calls get_ep_stat and register_metrics.
This code (before this patch) uses _active_ scheduling group to eventually add
metrics, using a local dict as guard against double regs. If, as described above,
we're called in a different sched group than the original one however, this
can cause double registrations.
Fixed here by keeping a reference to creating scheduling group and using this, not
active one, when/if creating new metrics.
Closes#14294
The metrics that are being deregistered (in this PR) cause Scylla to crash when a
table is dropped, but the corresponding table object in memory is not
yet deallocated, and a new table with the same name is created. This
caused a double-metrics-registration exception to be thrown. In order to
avoid it, we are deregistering table's metrics as soon as the table is
marked to be disposed from the database. Table's representation in memory can
still live, but shouldn't forbid other table with the same name to be
created.
Fixes#13548Closes#13971
This patch reduces the number of metrics ScyllaDB generates.
Motivation: The combination of per-shard with per-scheduling group
generates a lot of metrics. When combined with histograms, which require
many metrics, the problem becomes even bigger.
The two tools we are going to use:
1. Replace per-shard histograms with summaries
2. Do not report unused metrics.
The storage_proxy stats holds information for the API and the metrics
layer. We replaced timed_rate_moving_average_and_histogram and
time_estimated_histogram with the unfied
timed_rate_moving_average_summary_and_histogram which give us an option
to report per-shard summaries instead of histogram.
All the counters, histograms, and summaries were marked as
skip_when_empty.
The API was modified to use
timed_rate_moving_average_summary_and_histogram.
Closes#11173
Due to its sharded and token-based architecture, Scylla works best when the user workload is more or less uniformly balanced across all nodes and shards. However, a common case when this assumption is broken is the "hot partition" - suddenly, a single partition starts getting a lot more reads and writes in comparison to other partitions. Because the shards owning the partition have only a fraction of the total cluster capacity, this quickly causes latency problems for other partitions within the same shard and vnode.
This PR introduces per-partition rate limiting feature. Now, users can choose to apply per-partition limits to their tables of choice using a schema extension:
```
ALTER TABLE ks.tbl
WITH per_partition_rate_limit = {
'max_writes_per_second': 100,
'max_reads_per_second': 200
};
```
Reads and writes which are detected to go over that quota are rejected to the client using a new RATE_LIMIT_ERROR CQL error code - existing error codes didn't really fit well with the rate limit error, so a new error code is added. This code is implemented as a part of a CQL protocol extension and returned to clients only if they requested the extension - if not, the existing CONFIG_ERROR will be used instead.
Limits are tracked and enforced on the replica side. If a write fails with some replicas reporting rate limit being reached, the rate limit error is propagated to the client. Additionally, the following optimization is implemented: if the coordinator shard/node is also a replica, we account the operation into the rate limit early and return an error in case of exceeding the rate limit before sending any messages to other replicas at all.
The PR covers regular, non-batch writes and single-partition reads. LWT and counters are not covered here.
Results of `perf_simple_query --smp=1 --operations-per-shard=1000000`:
- Write mode:
```
8f690fdd47 (PR base):
129644.11 tps ( 56.2 allocs/op, 13.2 tasks/op, 49785 insns/op)
This PR:
125564.01 tps ( 56.2 allocs/op, 13.2 tasks/op, 49825 insns/op)
```
- Read mode:
```
8f690fdd47 (PR base):
150026.63 tps ( 63.1 allocs/op, 12.1 tasks/op, 42806 insns/op)
This PR:
151043.00 tps ( 63.1 allocs/op, 12.1 tasks/op, 43075 insns/op)
```
Manual upgrade test:
- Start 3 nodes, 4 shards each, Scylla version 8f690fdd47
- Create a keyspace with scylla-bench, RF=3
- Start reading and writing with scylla-bench with CL=QUORUM
- Manually upgrade nodes one by one to the version from this PR
- Upgrade succeeded, apart from a small number of operations which failed when each node was being put down all reads/writes succeeded
- Successfully altered the scylla-bench table to have a read and write limit and those limits were enforced as expected
Fixes: #4703Closes#9810
* github.com:scylladb/scylla:
storage_proxy: metrics for per-partition rate limiting of reads
storage_proxy: metrics for per-partition rate limiting of writes
database: add stats for per partition rate limiting
tests: add per_partition_rate_limit_test
config: add add_per_partition_rate_limit_extension function for testing
cf_prop_defs: guard per-partition rate limit with a feature
query-request: add allow_limit flag
storage_proxy: add allow rate limit flag to get_read_executor
storage_proxy: resultize return type of get_read_executor
storage_proxy: add per partition rate limit info to read RPC
storage_proxy: add per partition rate limit info to query_result_local(_digest)
storage_proxy: add allow rate limit flag to mutate/mutate_result
storage_proxy: add allow rate limit flag to mutate_internal
storage_proxy: add allow rate limit flag to mutate_begin
storage_proxy: choose the right per partition rate limit info in write handler
storage_proxy: resultize return types of write handler creation path
storage_proxy: add per partition rate limit to mutation_holders
storage_proxy: add per partition rate limit info to write RPC
storage_proxy: add per partition rate limit info to mutate_locally
database: apply per-partition rate limiting for reads/writes
database: move and rename: classify_query -> classify_request
schema: add per_partition_rate_limit schema extension
db: add rate_limiter
storage_proxy: propagate rate_limit_exception through read RPC
gms: add TYPED_ERRORS_IN_READ_RPC cluster feature
storage_proxy: pass rate_limit_exception through write RPC
replica: add rate_limit_exception and a simple serialization framework
docs: design doc for per-partition rate limiting
transport: add rate_limit_error
Adds a metric "read_rate_limited" which indicates how many times a read
operation was rejected due to per-partition rate limiting. The metric
differentiates between reads rejected by the coordinator and reads
rejected by replicas.
Adds a metric "write_rate_limited" which indicates how many times a
write operation was rejected due to per-partition rate limiting. The
metric differentiates between writes rejected by the coordinator and
writes rejected by replicas.
The reference is already at hand. The get_ep_stats() calls another
helper that also maps endpoint to datacenter, but it can get the
obtained dc sstring via argument.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The latter will need it to get dc info from. All the callers are either
storage proxy or have storage proxy pointer/reference to get topology
from.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Instead of lengthy blurbs, switch to single-line, machine-readable
standardized (https://spdx.dev) license identifiers. The Linux kernel
switched long ago, so there is strong precedent.
Three cases are handled: AGPL-only, Apache-only, and dual licensed.
For the latter case, I chose (AGPL-3.0-or-later and Apache-2.0),
reasoning that our changes are extensive enough to apply our license.
The changes we applied mechanically with a script, except to
licenses/README.md.
Closes#9937
When there are too many in-flight hints, writes start returning
overloaded exceptions. We're missing metrics for that, and these could
be useful when judging if the system is in overloaded state.
Paxos may leave an operation in a background after returning result to a
caller. Lest add a counter for background/foreground paxos handlers so
that it will be easier to detect memory related issues.
Message-Id: <20200510092942.GA24506@scylladb.com>
The .get_ep_stat(ep) call can throw when registering metrics (we have
issue for it, #5697). This is not expected by it callers, in particular
abstract_write_response_handler::timeout_cb breaks in the middle and
doesn't call the on_timeout() and the _proxy->remove_response_handler(),
which results in not removed and not released responce handler. In turn
not released response handler doesn't set the _ready future on which
response_wait() waits -> stuck.
Although the issue with .get_ep_stat() should be fixed, an exception in
it mustn't lead to deadlocks, so the fix is to make the get_ep_stat()
noexcept by catching the exception and returning a dummy stat object
instead to let caller(s) finish.
Fixes#5985
Tests: unit(dev)
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Message-Id: <20200430163639.5242-1-xemul@scylladb.com>
The learning stage of PAXOS protocol leaves behind an entry in
system.paxos table with the last learned value (which can be large). In
case not all participants learned it successfully next round on the same
key may complete the learning using this info. But if all nodes learned
the value the entry does not serve useful purpose any longer.
The patch adds another round, "prune", which is executed in background
(limited to 1000 simultaneous instances) and removes the entry in
case all nodes replied successfully to the "learn" round. It uses the
ballot's timestamp to do the deletion, so not to interfere with the
next round. Since deletion happens very close to previous writes it will
likely happen in memtable and will never reach sstable, so that reduces
memtable flush and compaction overhead.
Fixes#5779
Message-Id: <20200330154853.GA31074@scylladb.com>
Presently lightweight transactions piggy back the old
row value on prepare round response. If one of the participants
did not provide the old value or the values from peers don't match,
we perform a full read round which will repair the Paxos table and the
base table, if necessary, at all participants.
Capture the fact that read optimization has failed in a metric.
Message-Id: <20200304192955.84208-2-kostja@scylladb.com>
There is a case in current PAXOS implementation where timeout is
returned because the code cannot guaranty whether the value is accepted
or not in case of a contention. The counter will help to correlate this
condition with failed requests.
Message-Id: <20200211160653.30317-2-gleb@scylladb.com>
This commit builds on top of the introduced per scheduling group
statistics template and employs it for achieving a per scheduling
group statistics in storage_proxy.
Some of the statistics also had meaning as a global - per
shard one. Those are the ones for determining if to
throttle the write request. This was handled by creating a
global stats struct that will hold those stats and by changing
the stat update to also include the global one.
One point that complicated it is an already existing aggregation
over the per shard stats that now became a per scheduling group
per shard stats, converting the aggregation to a two-dimensional
aggregation.
One thing this commit doesn't handle is validating that an individual
statistic didn't "cross a scheduling group boundary", such validation
is possible but it can easily be added in the future. There is a
subtlety to doing so since if the operation did cross to other
scheduling group two connected statistics can lose balance
for example written bytes and completed write transactions.
Signed-off-by: Eliran Sinvani <eliransin@scylladb.com>
The storage proxy statistics structure did not contain
a method for registering the statistics for metric
groups, instead, each user had to register some
of the metrics by itself. There is no real reason
for separating the metrics registration from
the statistics data. There is even less justification
for doing this only for part of the stats as is
the case for those statistics.
This commit internalize the metrics registration
in the storage_proxy stats structures.
Signed-off-by: Eliran Sinvani <eliransin@scylladb.com>
This patch implements accounting of Cassandra's metrics related to
lightweight transactions, namely:
cas_read_latency transactional read latency (histogram)
cas_write_latency transactional write latency (histogram)
cas_read_timeouts number of transactional read timeouts
cas_write_timeouts number of transactional write timeouts
cas_read_unavailable number of transactional read
unavailable errors
cas_write_unavailable number of transactional write
unavailable errors
cas_read_unfinished_commit number of transaction commit attempts
that occurred on read
cas_write_unfinished_commit number of transaction commit attempts
that occurred on write
cas_write_condition_not_met number of transaction preconditions
that did not match current values
cas_read_contention how many contended reads were
encountered (histogram)
cas_write_contention how many contended writes were
encountered (histogram)
A read which arrived to a non-replica and had to be forwarded to a
replica by the coordinator is accounted in an own metric,
reads_coordinator_outside_replica_set.
Most often such read is produced by a driver which is unaware of
token distribution on the ring.
If a read was forwarded to another replica due to heat weighted
load balancing or query preference set by the user, it's not accounted
in the metric.
In case of a multi-partition read (a query using IN statement,
e.g. x in (1, 2, 3)), if any of the keys is read from a
non-local node the read is accounted as a non-local.
The rationale behind it is that if the user tries to be careful and send
IN queries only to the same vnode, they are rewarded with the counter
staying at zero, while if they send multi-partition IN queries without
any precautions, they will see the metric go up which gives them a
starting point for investigating performance problems.
Closes#4338
Add a metric to account writes which arrived to a non-replica and
had to be forwarded by a coordinator to a replica.
The name of the added metric is 'writes_coordinator_outside_replica_set'.
Do not account forwarded read repair writes, since they are already
accounted by a reads_coordinator_outside_replica_set metric, added in a
subsequent patch.
In scope of #4338.
The materialized views flow control mechanism works by adding a certain
delay to each client request, designed to slow down the client to the
rate at we can complete the background view work. Until now we could observe
this mechanism only indirectly, in whether or not it succeeded to keep the
view backlog bounded; But we had no way to directly observe the delay that
we decided to add. In fact, we had a bug where this delay was constantly
zero, and we didn't even notice :-)
So in this patch we add a new metric,
scylla_storage_proxy_coordinator_last_mv_flow_control_delay
The metric is a floating point number, in units of seconds.
This metric is somewhat peculiar that it always contains the *last* delay
used for some request - unlike other metrics it doesn't measure the "current"
value of something. Moreover, it can jump wildly because there is no
guarantee that each request's delay will be identical (in particular,
different requests may involve different base replicas which have different
view backlogs, so decide on different delays). In the future we may want
to supplement this metric with some sort of delay histogram. But even
this simple metric is already useful to debug certain scenarios and
understand if the materialized-views flow control is working or not.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20190227133630.26328-1-nyh@scylladb.com>
It is useful to have this counter to investigate the reason for read
repairs. Non zero value means that writes were lost after CL is reached
and RR is expected.
Message-Id: <20181009120900.GF22665@scylladb.com>
The foreground reads metric is derived from the number of live read
executors minus the number of background reads. Background reads are
counted down when their resolver times out. However, a read executor
may still be around for a while, resulting in such reads being
accounted as foreground.
Usually, the gap in which this happens is short, because executor
reference holders timeout quickly as well. It's not always the case
though. For instance, local read executor doesn't time out quickly
when the target shard has an overloaded CPU, and it takes a while
before the request goes through all the queues, even if IO is not
involved. Observed in #3628.
Fixes#3734.
Another problem is that all reads which received CL responses are
accounted as background, until all replicas respond, but if such read
needs reconciliation, it's still practically a foreground read and
should be accounted as such. Found during code review.
Fixes#3745.
This patch fixes both issues by rearranging accounting to track
foreground reads instead of background reads, and considering all
reads as foreground until the resulting promise is resolved.
Message-Id: <1535999620-25784-1-git-send-email-tgrabiec@scylladb.com>
Count operations which were started on one shard and
were performed on another, due to non-shard-aware driver
and/or RPC.
Message-Id: <20180723155118.8545-1-avi@scylladb.com>
This commit extracts metrics related to writes from stats structure,
so it can be easily replaced later, e.g. for materialized view metrics.
References #3385
References #3416
