During raft upgrade, a node may gossip about a new CDC generation that
was propagated through raft. The node that receives the generation by
gossip may have not applied the raft update yet, and it will not find
the generation in the system tables. We should consider this error
non-fatal and retry to read until it succeeds or becomes obsolete.
Another issue is when we fail with a "fatal" exception and not retrying
to read, the cdc metadata is left in an inconsistent state that causes
further attempts to insert this CDC generation to fail.
What happens is we complete preparing the new generation by calling `prepare`,
we insert an empty entry for the generation's timestamp, and then we fail. The
next time we try to insert the generation, we skip inserting it because we see
that it already has an entry in the metadata and we determine that
there's nothing to do. But this is wrong, because the entry is empty,
and we should continue to insert the generation.
To fix it, we change `prepare` to return `true` when the entry already
exists but it's empty, indicating we should continue to insert the
generation.
Fixesscylladb/scylladb#21227Closesscylladb/scylladb#22093
before this change, we rely on `using namespace seastar` to use
`seastar::format()` without qualifying the `format()` with its
namespace. this works fine until we changed the parameter type
of format string `seastar::format()` from `const char*` to
`fmt::format_string<...>`. this change practically invited
`seastar::format()` to the club of `std::format()` and `fmt::format()`,
where all members accept a templated parameter as its `fmt`
parameter. and `seastar::format()` is not the best candidate anymore.
despite that argument-dependent lookup (ADT for short) favors the
function which is in the same namespace as its parameter, but
`using namespace` makes `seastar::format()` more competitive,
so both `std::format()` and `seastar::format()` are considered
as the condidates.
that is what is happening scylladb in quite a few caller sites of
`format()`, hence ADT is not able to tell which function the winner
in the name lookup:
```
/__w/scylladb/scylladb/mutation/mutation_fragment_stream_validator.cc:265:12: error: call to 'format' is ambiguous
265 | return format("{} ({}.{} {})", _name_view, s.ks_name(), s.cf_name(), s.id());
| ^~~~~~
/usr/bin/../lib/gcc/x86_64-redhat-linux/14/../../../../include/c++/14/format:4290:5: note: candidate function [with _Args = <const std::basic_string_view<char> &, const seastar::basic_sstring<char, unsigned int, 15> &, const seastar::basic_sstring<char, unsigned int, 15> &, const utils::tagged_uuid<table_id_tag> &>]
4290 | format(format_string<_Args...> __fmt, _Args&&... __args)
| ^
/__w/scylladb/scylladb/seastar/include/seastar/core/print.hh:143:1: note: candidate function [with A = <const std::basic_string_view<char> &, const seastar::basic_sstring<char, unsigned int, 15> &, const seastar::basic_sstring<char, unsigned int, 15> &, const utils::tagged_uuid<table_id_tag> &>]
143 | format(fmt::format_string<A...> fmt, A&&... a) {
| ^
```
in this change, we
change all `format()` to either `fmt::format()` or `seastar::format()`
with following rules:
- if the caller expects an `sstring` or `std::string_view`, change to
`seastar::format()`
- if the caller expects an `std::string`, change to `fmt::format()`.
because, `sstring::operator std::basic_string` would incur a deep
copy.
we will need another change to enable scylladb to compile with the
latest seastar. namely, to pass the format string as a templated
parameter down to helper functions which format their parameters.
to miminize the scope of this change, let's include that change when
bumping up the seastar submodule. as that change will depend on
the seastar change.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Before this patch, writes to the previous CDC generations would
always be rejected. After this patch, they will be accepted if
the write's timestamp is greater than `now - generation_leeway`.
This change was proposed around 3 years ago. The motivation was
to improve user experience. If a client generates timestamps by
itself and its clock is desynchronized with the clock of the node
the client is connected to, there could be a period during
generation switching when writes fail. We didn't consider this
problem critical because the client could simply retry a failed
write with a higher timestamp. Eventually, it would succeed. This
approach is safe because these failed writes cannot have any side
effects. However, it can be inconvenient. Writing to previous
generations was proposed to improve it.
The idea was rejected 3 years ago. Recently, it turned out that
there is a case when the client cannot retry a write with the
increased timestamp. It happens when a table uses CDC and LWT,
which makes timestamps permanent. Once Paxos commits an entry with
a given timestamp, Scylla will keep trying to apply that entry
until it succeeds, with the same timestamp. Applying the entry
involves writing to the CDC log table. If it fails, we get stuck.
It's a major bug with an unknown perfect solution.
Allowing writes to previous generations for `generation_leeway` is
a probabilistic fix that should solve the problem in practice.
Note that allowing writes only to the previous generation might
not be enough. With the Raft-based topology, it is possible to
add multiple nodes concurrently. Moreover, tablets make streaming
instant, which allows the topology coordinator to add multiple nodes
very quickly. So, creating generations with almost identical
timestamps is possible. Then, we could encounter the same bug but,
for example, for a generation before the previous generation.
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
Currently, whole topology description for CDC is stored in a single row.
This means that for a large cluster of strong machines (say 100 nodes 64
cpus each), the size of the topology description can reach 32MB.
This causes multiple problems. First of all, there's a hard limit on
mutation size that can be written to Scylla. It's related to commit log
block size which is 16MB by default. Mutations bigger than that can't be
saved. Moreover, such big partitions/rows cause reactor stalls and
negatively influence latency of other requests.
This patch limits the size of topology description to about 4MB. This is
done by reducing the number of CDC streams per vnode and can lead to CDC
data not being fully colocated with Base Table data on shards. It can
impact performance and consistency of data.
This is just a quick fix to make it easily backportable. A full solution
to the problem is under development.
For more details see #7961, #7993 and #7985.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Refs #6864
When booting a clean scylla, CDC stream ID:s will not be availble until
a n*ring delay time period has passed. Before this, writing to a CDC
enabled table will fail hard.
For alternator (and its tests), we can report the stream(s) for tables as not yet
available (ENABLING) until such time as id:s are
computed.
v2:
* Keep storage service ref in executor
CDC can get all it needs from a config and does not need
partitioner.
For base table specific operations CDC is using partitioner
from that table (obtained with schema::get_partitioner).
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Since token representation is fixed now, all the partitioners
will share the sharding logic. It makes sense now to keep
the logic in common super class and separate header that's
included only in i_partitioner.cc.
shard_of and token_for_next_shard are now implemented in
i_partitioner. They would be non-virtual but we have to
keep them virtual because one test is overriding them
to enforce some specific sharding.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
The class stores a queue of CDC generations to be used for choosing
streams when writing to the CDC log.
This data structure will be updated on some gossip events (when a new node
joins the cluster and proposes a new generation of CDC streams).
