The restart logic is wrong because C* had a bug in
bf599fb5b062cbcc652da78b7d699e7a01b949ad and they fixed later and we
translated the broken version. We must check if there is an existing
endpoint state and call on_restart() hooks on that, not the newly
available endpoint state.
Spotted while inspecting the code.
Acked-by: Asias He <asias@scylladb.com>
From Avi:
Memtables do not use an allocating_section to guard against allocation
failure, and hence can fail an allocation. Reproducible by changing
perf_mutation to use an allocating type (bytes_type with a nontrivial
size) and making the loop longer.
Fix by using an allocating_section.
Recently, I have introduced cf_stats into the database, propagating all the way
back to the column family. The problem, however, is that some tests create a
column family config themselves instead of going through make_column_family.
That is ultimately ok if those tests are not expected to flush memtables. But
if they are, the cf_stats pointer will be null and we will crash. Although
there are many solutions to this, the one that is in tune with our current
practices is to have the test that requires it provide an empty cf_stats storage
area that can be written to. That's already how we handle the disk directory and
other things like compaction properties.
With this patch, test.py passes again.
Signed-off-by: Glauber Costa <glommer@scylladb.com>
This patch substitutes uint64_t for uint32_t as the type for
commitlog_total_space_in_mb. Moving to 64 is not strictly needed, since even a
signed 32-bit type would allow us to easily handle 2TB. But since we store that
in the commitlog as a 64-bit value, let's match it.
Moving from unsigned to signed, however, allow us to represent negative
numbers. With that in place, we can change the semantics of the value
slightly, so to allow a negative number to mean "all memory".
The reason behind this, is that the default value "8GB", is an artifact of the
JVM. We don't need that, and in many-shards configuration, each shard flushes
the commitlog way too often, since 8GB / many_shards = small_number.
8GB also happens to be a popular heap size for C* in the JVM. For us, we would
like to equate that (at least) with the amount of memory. The problem is how to
do that without introducing new options or changing the semantics of existing
options too radically.
The proposed solution will allow us to still parse C* yaml files, since those
will always have positive numbers, while introducing our own defaults.
Signed-off-by: Glauber Costa <glommer@scylladb.com>
Debian package system has two types of package, 'native' and 'non-native'.
'native' is the package just for Debian, it contains debian/ directory source tar.gz, doesn't have debian.tar.gz.
'non-native' has orig.tar.gz which is upstream source code tar ball, then it has debian.tar.gz which contains debian/ directory.
Scylla is 'native' now but should be 'non-native' since this is not just for Debian, so move debian/ to dist/ubuntu/, make orig.tar.gz using git-archive-all, copy dist/ubuntu/debian/ to debian/ then generate debian.tar.gz.
atomic_cell will soon become type-aware, so add helpers to class operation
that can supply the type, as it is available in operation::column.type.
(the type will be used in following patches)
schema_tables manages some boolean columns stored in system tables; it
dynamically creates them from C++ values. But as we lacked bool->data_value
conversion, the C++ value was converted to a int32_type. Somehow this didn't
cause any problems, but with some pending patches I have, it does.
Add a bool->data_value converting constructor to fix this.
Since bytes is a very generic value that is returned from many calls,
it is easy to pass it by mistake to a function expecting a data_value,
and to get a wrong result. It is impossible for the data_value constructor
to know if the argument is a genuine bytes variable, a data_value of another
type, but serialized, or some other serialized data type.
To prevent misuse, make the data_value(bytes) constructor
(and complementary data_value(optional<bytes>) explicit.
When do_stop_native_transport exits, cserver is destroyed which can
happen before cserver->stop(). Fix by capturing cserver in
cserver->stop()'s continuation to extend its lifetime. The same for
thrift server.
scylla: scylla/seastar/core/sharded.hh:327: seastar::sharded<Service>::~sharded()
[with Service = transport::cql_server]: Assertion `_instances.empty()' failed.
When analyzing a recent performance issue, I found helpful to keep track of
the amount of memtables that are currently in flight, as well as how much memory
they are consuming in the system.
Although those are memtable statistics, I am grouping them under the "cf_stats"
structure: being the column family a central piece of the puzzle, it is reasonable
to assume that a lot of metrics about it would be potentially welcome in the future.
Note that we don't want to reuse the "stats" structure in the column family: for once,
the fields not always map precisely (pending flushes, for instance, only tracks explicit
flushes), and also the stats structure is a lot more complex than we need.
Signed-off-by: Glauber Costa <glommer@scylladb.com>
* seastar 5c10d3e...20bf03b (5):
> do not re-throw exception to get to an exception pointer
> Adding timeout counter to the rpc
> configure.py: support for pkg-config before release 0.28
> future: don't forget to warn about ignored exception
> tutorial: continue network API section
Found by debug build
==10190==ERROR: AddressSanitizer: new-delete-type-mismatch on 0x602000084430 in thread T0:
object passed to delete has wrong type:
size of the allocated type: 16 bytes;
size of the deallocated type: 8 bytes.
#0 0x7fe244add512 in operator delete(void*, unsigned long) (/lib64/libasan.so.2+0x9a512)
#1 0x3c674fe in std::default_delete<dht::range_streamer::i_source_filter>::operator()(dht::range_streamer::i_source_filter*)
const /usr/include/c++/5.1.1/bits/unique_ptr.h:76
#2 0x3c60584 in std::unique_ptr<dht::range_streamer::i_source_filter, std::default_delete<dht::range_streamer::i_source_filter> >::~unique_ptr()
/usr/include/c++/5.1.1/bits/unique_ptr.h:236
#3 0x3c7ac22 in void __gnu_cxx::new_allocator<std::unique_ptr<dht::range_streamer::i_source_filter,
std::default_delete<dht::range_streamer::i_source_filter> > >::destroy<std::unique_ptr<dht::range_streamer::i_source_filter,
std::default_delete<dht::range_streamer::i_source_filter> > >(std::unique_ptr<dht::range_streamer::i_source_filter,
std::default_delete<dht::range_streamer::i_source_filter> >*) /usr/include/c++/5.1.1/ext/new_allocator.h:124
...
Fixes#549.
Being clinically absent-minded, aggregate query support (i.e. count(...))
was left out of the "paging" change set.
This adds repeated paged querying to do aggregate queries (similar to
origin). Uses "batched" paging.
For compatibility reasons, compaction_strategy should accept both class
name strategy and the full class name that includes the package name.
In origin the result name depends on the configuration, we cannot mimic
that as we are using enum for the type.
So currently the return class name remains the class itself, we can
consider changing it in the future.
If the name is org.apache.cassandra.db.compaction.Name the it will be
compare as Name
The error message was modified to report the name it was given.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Fixes#545
"Slight file format change for commitlog segments, now incluing
a scylla "marker". Allows for fast-fail if trying to load an
Origin segment.
WARNING: This changes the file format, and there is no good way for me to
check if a CL is "old" scylla, or Origin (since "version" is the same). So
either "old" scylla files also fail, or we never fail (until later, and
worse). Thus, if upgrading from older to this patch ensure to
have cleaned out all commit logs first."
Fixes#355
"Implements query paging similar to origin. If driver sets a "page size" in
a query, and we cannot know that we will not exceed this limit in a single
query, the query is performed using a "pager" object, which, using modified
partition ranges and query limits, keeps track of returned rows to "page"
through the results.
Implementation structure sort of mimics the origin design, even though it
is maybe a little bit overkill for us (currently). On the other hand, it
does not really hurt.
This implementation is tested using the "paging_test" subset in dtest.
It passes all test except:
* test_paging_using_secondary_indexes
* test_paging_using_secondary_indexes_with_static_cols
* test_failure_threshold_deletions
The two first because we don't have secondary indexes yet, the latter
because the test depends on "tombstone_failure_threshold" in origin.
Potential todo: Currently the pager object does not shortcut result
building fully when page limit is exceeded. Could save a little work
here, but probably not very significant."
Allows us fail fast if someone tries to replay an Origin commit log.
WARNING: This changes the file format, and there is no good way for me to
check if a CL is "old" scylla, or Origin (since "version" is the same). So
either "old" scylla files also fail, or we never fail (until later, and
worse). Thus, if upgrading from older to this patch, likewise, ensure to
have cleaned out all commit logs first.
