This patch implements the size_estimates_recorder, which periodically
writes estimations for all the non-system column families in the
size_estimates system table. The size_estimates_recorder class
corresponds to the one in Cassandra's SizeEstimatesRecorder.java.
Estimation is carried out by shard 0. Since we're estimating based on
data in shared sstables, having multiple shards doing this would skew
the results.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch makes hashing for repair calculate checksums in a way that
doesn't require rebuilding whole mutation.
Unfortunately, such checksums are incompatible with the old ones so the
old way for computing checksums is preserved for compatibility reasons.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
At the moment, we only trigger compaction after creating a new
sstable as a result of memtable flush, or some other event such
as changing compaction strategy of a column family.
However, it's important to trigger compaction on boot too.
That will happen after loading all column families.
Fixes#1404.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <54d38a418157454eec97aaba6b8a6b6e51484db4.1467135349.git.raphaelsc@scylladb.com>
If we don't, std::terminate() causes a core dump, even though an
exception is sort-of-expected here and can be handled.
Add an exception handler to fix.
Fixes#1379.
Message-Id: <1466595221-20358-1-git-send-email-avi@scylladb.com>
After commit faa4581, each shard only starts splitting its shared sstables
after opening all sstables. This was important because compaction needs to
be aware of all sstables.
However, another bug remained: If one shard finishes loading its sstables
and starts the splitting compactions, and in parallel a different shard is
still opening sstables - the second shard might find a half-written sstable
being written by the first shard, and abort on a malformed sstable.
So in this patch we start the shared sstable rewrites - on all shards -
only after all shards finished loading their sstables. Doing this is easy,
because main.cc already contains a list of sequential steps where each
uses invoke_on_all() to make sure the step completes on all shards before
continuing to the next step.
Fixes#1371
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <1466426641-3972-1-git-send-email-nyh@scylladb.com>
"Reclaiming many segments was observed to cause up to multi-ms
latency. With the new setting, the latency of reclamation cycle with
full segments (worst case mode) is below 1ms.
I saw no difference in throughput in a CQL write micro benchmark
in neither of these workloads:
- full segments, reclaim by random eviction
- sparse segments (3% occupancy), reclaim by compaction and no eviction
Fixes #1274."
tracing::tracing local instance is dereferenced from a
cql_server::connection::process_request(), therefore tracing::tracing
service may be stop()ed only after a CQL server service is down.
On the other hand it may not be stopped before RPC service is down
because a remote side may request a tracing for a specific command too.
This patch splits the tracing::tracing stop() into two phases:
1) Flush all pending tracing records and stop the backend.
2) Stop the service.
The first phase is called after CQL server is down and before RPC is down.
The second phase is called after RPC is down.
Fixes#1339
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Message-Id: <1465840496-19990-1-git-send-email-vladz@cloudius-systems.com>
Since messaging_service is fully initialized in
storage_service::init_server which calls
messaging_service::start_listen, we need to delay
the messaging_service api registration after it.
This reverts commit b3ed55be1d.
The issue is in the failing dtest, not this commit. Gleb writes:
"The bug is in the test, not the patch. Test waits for repair session
to end one way or the other when node is killed, but for nodetool to
know if repair is completed it needs to poll for it. If node dies
before nodetool managed to see repair completion it will stuck
forever since jmx is alive, but does not provide answers any more.
The patch changes timing, repair is completed much close to exit now,
so problem appears, but it may happen even without the patch.
The fix is for dtest to kill jmx as part of killing a node
operation."
Now that Lucas fixed the problem in scylla-ccm, revert the revert.
compact_on_idle will lead users to thinking we're talking about sstable
compaction, not log-structured-allocator compaction.
Rename the variable to reduce the probability of confusion.
Message-Id: <1464261650-14136-1-git-send-email-avi@scylladb.com>
Currently, we register snitch API in set_server_gossip_settle() which
waits until a node has joined the cluster. This makes 'nodetool status'
not properly show the status of a joining node. Fix the issue by
registering snitch API earlier.
Fixes#1269.
Message-Id: <1463576381-15484-1-git-send-email-penberg@scylladb.com>
Avi says:
"During shutdown, we prevent new compactions, but perhaps too late.
Memtables are flushed and these can trigger compaction."
To solve that, let's stop compaction manager at a very early step
of shutdown. We will still try to stop compaction manager in
database::stop() because user may ask for a shutdown before scylla
was fully started. It's fine to stop compaction manager twice.
Only the first call will actually stop the manager.
Fixes#1238.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <c64ab11f3c91129c424259d317e48abc5bde6ff3.1462496694.git.raphaelsc@scylladb.com>
A shared sstable must be compacted by all shards before it can be deleted.
Since we're stoping, that's not going to happen. Cancel those pending
deletions to let anyone waiting on them to continue.
While Seastar in general can accept any parameter for its I/O queues, Scylla
in particular shouldn't run with them disabled. Such will be the status when
the max-io-requests parameter is not enabled.
On top of that, we would like to have enough depth per I/O queue not to allow
for shard-local parallelism. Therefore, we will require a minimum per-queue
capacity of 4. In machines where the disk iodepth is not enough to allow for 4
concurrent requests per shard, one should reduce the number of I/O queues.
For --max-io-requests, we will check the parameter itself. However, the
--num-io-queues parameter is not mandatory, and given enough concurrent
requests, Seastar's default configuration can very well just be doing the right
thing. So for that, we will check the final result of each I/O queue.
As it is the case with other checks of the sorts, this can be overridden by
the --developer-mode switch.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <63bf7e91ac10c95810351815bb8f5e94d75592a5.1458836000.git.glauber@scylladb.com>
Since initialization now runs in a thread storage, messaging and
gossiper services initialization code may take advantage of it too.
Message-Id: <20160323094732.GF2282@scylladb.com>
Fix the validation error message to look like this:
Scylla version 666.development-20160316.49af399 starting ...
WARN 2016-03-17 12:24:15,137 [shard 0] config - Option partitioner is not (yet) used.
WARN 2016-03-17 12:24:15,138 [shard 0] init - NOFILE rlimit too low (recommended setting 200000, minimum setting 10000; you may run out of file descriptors.
ERROR 2016-03-17 12:24:15,138 [shard 0] init - Bad configuration: invalid 'listen_address': eth0: boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::system::system_error> > (Invalid argument)
Exiting on unhandled exception of type 'bad_configuration_error': std::exception
Instead of:
Exiting on unhandled exception of type 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::system::system_error> >': Invalid argument
Fixes#1051.
Message-Id: <1458210329-4488-1-git-send-email-penberg@scylladb.com>
Defer registering services to the API server until commitlog has been
replayed to ensure that nobody is able to trigger sstable operations via
'nodetool' before we are ready for them.
Message-Id: <1458116227-4671-1-git-send-email-penberg@scylladb.com>
Streaming is used by bootstrap and repair. Streaming uses storage_proxy
class to apply the frozen_mutation and db/column_family class to
invalidate row cache. Defer the initalization just before repair and
bootstrap init.
Message-Id: <8e99cf443239dd8e17e6b6284dab171f7a12365c.1458034320.git.asias@scylladb.com>
Register the REPAIR_CHECKSUM_RANGE messaging service verb handler after
we have replayed the commitlog to avoid responding with bogus checksums.
Message-Id: <1458027934-8546-1-git-send-email-penberg@scylladb.com>
Defer registering migration manager RPC verbs after commitlog has has
been replayed so that our own schema is fully loaded before other other
nodes start querying it or sending schema updates.
Message-Id: <1457971028-7325-1-git-send-email-penberg@scylladb.com>
Deletion of previous stale, temporary SSTables is done by Shard0. Therefore,
let's run Shard0 first. Technically, we could just have all shards agree on the
deletion and just delete it later, but that is prone to races.
Those races are not supposed to happen during normal operation, but if we have
bugs, they can. Scylla's Github Issue #1014 is an example of a situation where
that can happen, making existing problems worse. So running a single shard
first and getting making sure that all temporary tables are deleted provides
extra protection against such situations.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
While looking at initialization code I felt like my head is going to
explode. Moving initialization into a thread makes things a little bit
better. Only lightly tested.
Message-Id: <20160310163142.GE28529@scylladb.com>
We start services like gossiper before system keyspace is initialized
which means we can start writing too early. Shuffle code so that system
keyspace is initialized earlier.
Refs #1014
Message-Id: <1457593758-9444-1-git-send-email-penberg@scylladb.com>
We require SSE 4.2 (for commitlog CRC32), verify it exists early and bail
out if it does not.
We need to check early, because the compiler may use newer instructions
in the generated code; the earlier we check, the lower the probability
we hit an undefined opcode exception.
Message-Id: <1456665401-18252-1-git-send-email-avi@scylladb.com>
When shutting down a node gracefully, this patch asks all ongoing repairs
started on this node to stop as soon as possible (without completing
their work), and then waits for these repairs to finish (with failure,
usually, because they didn't complete).
We need to do this, because if the repair loop continues to run while we
start destructing the various services it relies on, it can crash (as
reported in #699, although the specific crash reported there no longer
occurs after some changes in the streaming code). Additionally, it is
important that to stop the ongoing repair, and not wait for it to complete
its normal operation, because that can take a very long time, and shutdown
is supposed to not take more than a few seconds.
Fixes#699.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <1455218873-6201-1-git-send-email-nyh@scylladb.com>
Fixes#868.
Registerring exit hooks while reactor is already iterating over exit
hooks is not allowed and currently leads to undefined behavior
observed in #868. While we should make the failure more user friendly,
registering exit hooks concurrently with shutdown will not be allowed.
We don't expect exit hooks to be registered after exit starts because
this would violate the guarantee which says that exit hooks are
executed in reverse order of registration. Starting exit sequence in
the middle of initialization sequence would result in use after free
errors. Btw, I'm not sure if currently there's anything which prevents
this
To solve this problem, move the exit hook to initilization
sequence. In case of tests, the cleanup has to be called explicitly.
supervisor_notify() calls periodically, to log message on systemd.
So raise(SIGSTOP) will called multiple times, upstart doesn't expected that.
We need to call it just one time.
Fixes#846
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
We register engine().at_exit() callbacks when we initialize the services. We
do not really call the callbacks at the moment due to #293.
It is pretty hard to see the whole picture in which order the services
are shutdown. Instead of for each services to register a at_exit()
callbacks, I proposal to have a single at_exit() callback which do the
shutdown for all the services. In cassandra, the shutdown work is done
in storage_service::drain_on_shutdown callbacks.
In this patch, the drain_on_shutdown is executed during shutdown.
As a result, the proper gossip shutdown is executed and fixes#790.
With this patch, when Ctrl-C on a node, it looks like:
INFO [shard 0] storage_service - Drain on shutdown: starts
INFO [shard 0] gossip - Announcing shutdown
INFO [shard 0] storage_service - Node 127.0.0.1 state jump to normal
INFO [shard 0] storage_service - Drain on shutdown: stop_gossiping done
INFO [shard 0] storage_service - CQL server stopped
INFO [shard 0] storage_service - Drain on shutdown: shutdown rpc and cql server done
INFO [shard 0] storage_service - Drain on shutdown: shutdown messaging_service done
INFO [shard 0] storage_service - Drain on shutdown: flush column_families done
INFO [shard 0] storage_service - Drain on shutdown: shutdown commitlog done
INFO [shard 0] storage_service - Drain on shutdown: done
The API needs to be available at an early stage of the initialization,
on the other hand not all the specific APIs are available at that time.
This patch breaks the API initialization into stages, in each stage
additional commands will be available.
While setting that the api header files was broken into api_init.hh that
is relevent to the main and to api.hh which holds the different
api helper functions.
Fixes#754
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1453822331-16729-2-git-send-email-amnon@scylladb.com>
This reverts commit f0d68e4 ("main: start the http server in the first
step"). The service layer is not ready to serve clients before it's
fully up and running which causes early startup crashes everywhere.
Message-Id: <1452768015-22763-1-git-send-email-penberg@scylladb.com>
With Docker we might be running on a filesystem that does not support DMA
(aufs; or tmpfs on boot2docker), so let --developer-mode allow running
on those file systems.
Message-Id: <1452593083-25601-1-git-send-email-avi@scylladb.com>
