key() will not be valid for dummy entries, but position() is always
valid.
[tgrabiec: Extracted from other commits]
[tgrabiec: Added missing change to range_tombstone_stream::get_next]
This is a position that's always in the end after any
other position. It will be used for dummy rows_entry.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
It can happen that touch() will trigger eviction on entry to
allocating section, and drop in occupancy around insertion
will not happen. As a result, we may evict a lot without detecting that.
Extend the check to include touch() and use more reliable eviction counters.
We support situations where `seastar.pc` is available, and when it isn't.
When `seastar.pc` is available, we grab the compilation flags from it in
addition to the defaults.
Some DPDK files are statically available in the source repository. We prefer
those to files placed during compilation in case modifications are made during
development (that would be lost during a build).
We always disable GCC6 concepts for the IDE, even if they are enabled while
configuring the project for compilation. CLion's parser doesn't understand them.
One final benefit to this revision is that now only target-specific flags are
modified rather than global flags.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <75457652201c5ed89d05081ec4b56b4340721cf5.1498237756.git.jhaberku@scylladb.com>
"This patch series makes two noteworthy changes:
- `gc_clock` now uses `seastar::lowres_system_clock` for better performance.
- Code common to all clocks is properly refactored and shared.
Otherwise, there are some small improvements that should have no functional impact."
* 'jhk/lowres_gc_clock/v1' of https://github.com/hakuch/scylla:
Seal clock definitions
`timestamp_clock::now()` is not `noexcept`
Make `db_clock` `time_t` conversions `constexpr`
Move common clock implementation helpers
Simplify clock implementations
db_clock.hh: Clean preprocessor directives
Make `gc_clock` a model of `Clock`
Use `lowres_system_clock` to back `gc_clock`
Add `time_t` conversions for `gc_clock`
The problem is that `std::chrono::duration_cast` is not `noexcept`. As a result,
`timestamp_clock` is actually a model of `Clock` and not `TrivialClock`.
This change fixes the dependencies between the clock implementation headers. All
the clocks share the common clock offset, but are otherwise independent (though
the `db_clock` does depend on `gc_clock` for time point conversions).
`gc_clock` reports system time, and these conversion functions allow for
manipulating time points produced by the clock without making assumptions about
its epoch.
"Fix minor issues found while building with clang trunk."
* 'clang' of https://github.com/avikivity/scylla:
seastarx: don't make seastar namespace inline
seastarx: add missing make_shared forward declaration
tests: fix call to seastar::sleep()
dht: fix bad to_sstring() call
* 'lcs_improvements_v1' of github.com:raphaelsc/scylla:
lcs: remove useless code for choosing L0 candidates
lcs: remove some dead code
lcs: make logger static
lcs: actually prefer oldest sstables of L0 when it falls behind
lcs: remove useless expensive check for overlapping L1 sstables
so now user can look at nodetool compactionstats and determine
whether or not resharding is running, for example:
$ ./bin/nodetool compactionstats
pending tasks: 3
id compaction type keyspace table completed total unit progress
<none> RESHARD system compaction_history 11 256 keys 4.30%
<none> RESHARD system compaction_history 2 256 keys 0.78%
<none> RESHARD system compaction_history 10 256 keys 3.91%
<none> RESHARD system compaction_history 8 256 keys 3.12%
<none> RESHARD system compaction_history 7 256 keys 2.73%
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170620175733.25882-1-raphaelsc@scylladb.com>
The code filters CFs by name to not include system keyspace, but v3
schema added yet another system namespace. Better filter according to
replication strategy to accommodate for schema v4 adding even more
system keyspaces.
Fixes: #2516
Message-Id: <20170621073816.GB3944@scylladb.com>
This version optionally reads the include paths for Seastar from pkg-config and
uses file globbing to register all source and header files.
In comparison to the previous version, I see all files in the project explorer
view are "active" (rather than just .cc files). I believe there are also fewer
errors reported by the editor.
Add support Debian new stable release.
Also including following changes:
- update libthrift due to unable to compile on Debian 9
- drop dist/debian/supported_release since distribution check code moved to pbuilderrc
- add libssl-dev for build-depends
- add sudo for pbuilder extra packages (Debian doesn't have it by default install)
Signed-off-by: syuu <syuu@dokukino.com>
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1498047515-19972-1-git-send-email-syuu@scylladb.com>
The memtable destructor can take a long time if the memtable is full; use
clear_gently() to clear it without impacting latency.
Fixes#2477.
Message-Id: <20170620093550.16121-1-avi@scylladb.com>
Although usually one can fast_forward_to() on the result of a
sstable::as_mutation_source(), earlier we had an optimization
where if a single partition was specified, it was read exactly,
and fast_forward_to() was *NOT* allowed.
With the mutation_reader::forwarding flag patch, when this flag
was on - requesting fast_forward_to() - we disabled this optimization.
This makes sense, but is not backward compatible with the code which
previously assumes this optimization exists... So this patch returns
this optimization, despite this meaning that we blatently ignore
the fwd_mr flag in that case.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170620081107.14335-1-nyh@scylladb.com>
* seastar 621b7ed...9e2b7ec (8):
> Merge "Low-resolution clocks" from Jesse
> build: disable -Wattributes when gcc -fvisibility=hidden bug strikes
> build: work around ragel 7 generated code bug
> rpc: make unmarshall_exception() inline
> future-utils: make functions global
> fix reactor stall detector rate limiting on an mostly idle system
> prometheus: Add ability to add /metrics to any http_server
> prometheus: fix memory leak in http_server_control
multi_range_mutation_reader uses fast_forward_to() to skip between
ranges, so we always need to create the underlying reader with with
mutation_reader::forwarding::yes if there is more than one range,
irrespective of whether multi_range_mutation_reader itself will be
forwarded or not.
Fixes#2510.
Introduced in commit 3018df1.
Message-Id: <1497943032-18696-1-git-send-email-tgrabiec@scylladb.com>
The code being removed could be used if parallel compaction were
allowed for LCS, but the current code isn't even allowing that.
At the moment, it's only wasting cycles.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Strategy prefers promoting oldest sstables in L0. Because sort
procedure is incorrectly sorting elements in descending order,
newest sstables will be promoted first *if and only if* L0 falls
behind (more than 32 sstables). If L0 doesn't fall behind, we'll
have all L0 sstables compacted with overlapping ones in L1.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
there's no way a L1 sstable will be in candidates set which was
previously built from list of L0 sstables.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
