This commit defines two traits: `Con` and `ConOps`. Implementors of
`ConOps` get a free implementation for `Con`. `Con` is the ultimate
object that can be used in place of the current SSL/non-SSL connection
objects. If you look at the implementations of the current connection
objects, they have a lot of repetition as they do almost the same thing
except for the fact that they have a different underlying stream.
This is exactly what we're trying to eliminate. We will also define a
generic connection handler object to reduce redundancy.
Several changes were made to accomodate for this, including the addition
of the write_to_disk function that should be used by fns which don't
have a FileLock to pass for flushing data to the disk.
BGSAVE now takes ownership of a FileLock object which it uses for
running BGSAVE.
It seems that on Windows unlocking errors if the file has already been
unlocked. To fix this, we've added a platform-specific field to see if
the FileLock object has already been used to unlock the file.
This is the unlock field. In the Drop impl for Windows, we check the
unlocked flag to determine if we need to unlock the file.
This commit implements file locks for unix-based systems and windows
systems. This is done by using platform-specific `__sys` modules for
locking, trying to lock and unlocking files.
A build script was added for unix-systems that make use of the
flock-posix.c file
This commit removes the fscposix.c file and begins implementing native
file locking mechanisms for each platform (supported platforms)
BSD-style `flock`s were added
This commit adds a basic implementation of POSIX advisory record locking
which sets a lock on the `data.bin` file when the database server starts
and releases the lock when it terminates. This is just done for
compliance to let other processes know that we don't want them to use
the file.
However, the result depends entirely on the process that wants to do
'something' with the file. It is the responsibility of the process to
ensure that it respects the file lock.
Also, exclusive locks aren't perfect on Linux, so we can't rely on them.
See discussion #123 for more information.
This variant is absolutely redundant as we're just ignoring the entire
buffer and not just a part of it
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This closes#107, closes#108 and closes#109.
The configuration template was updated to include TLS/SSL and the
corresponding tests were also updated.
It also renames `sdb` to `skyd` for streamlining binary names.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit uses a Regex match iterator along with a few replace
operations to enable the parsing of quoted strings from arguments.
Previously, we simply ran a `split_whitespace()` to get the parts of the
ActionGroup, but now we're using this new Regex which enables arguments
like: 'SET me "sayan spaced"' to be passed and validated.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit now checks if the second value passed to MKSNAP points
to any parent/root directory before performing any action.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This call is necessary as SUPDATE returns Nil even if one of the keys
don't exist.
Also, this was note added to the actiondoc.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Just like fd139a9dda, this skip is not
needed as we're already breaking from the loop.
Also, all remaining unsafe blocks that were left in
2e85fbb831 have been explained.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
The skipping of the next value is absolutely unneeded as we're already
exiting the loop when the hash table contains the key.
Thus, this op was removed.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Dependencies were upgraded
A mirror sync badge was added. Also README was revised and workflow badge was fixed.
Signed-off-by: Sayan Nandan nandansayan@outlook.com
`openssl-sys` isn't required anymore; it was added in the `ssl` branch
during the development phase.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Also, dependencies were upgraded across all crates and the version for
`tdb-macros` was streamlined to 0.5.0 like the other crates.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
The data was being delievered in different batches, which
caused problems. This commit replaces the current stream
writer with a buffered writer ensuring good delivery.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
We were doing an extremely erroneous thing: writing to the TCP
socket instead of the SSL socket. This caused OpenSSL to report
problems on the client and server sides, telling us that there
was a problem with the SSL connection.
This commit revises the `write_lowlevel` trait impl for `SslStream` to
write to the SSL socket.
Also, the 'wrong' flushing of data for similar reasons has been fixed
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit enables queries to be executed on secure connections.
At the same time, the `execute_query_ssl` function was removed as
`execute_query` has been modified so that it can be used by both secure
and insecure connections
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit enables SSL settings to be read from command-line arguments
and from the configuration file.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This object either holds a mutable reference to an unencrypted TCP
stream or an encrypted TLS stream (using OpenSSL). The action handlers
were modified as a consequence.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit adds a basic SSL/TLS listener using `openssl`.
The `SslListener` object can accept a connection and get a decrypted
stream.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This makes sure that we can implement serialization for multiple types.
At the same time, we DON'T put in blanket implementations for this trait
as we want to make sure that unintended things don't get flushed.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
We've made the error handling slightly 'less aggressive' than the
previous version which `unwrap`ped here and there.
Also, the documentation for the entire snapstore module was greatly
improved.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This commit provides an implementation which allows multiple 'named'
namespaces to be serialized and deserialized.
No information about the data needs to be known for deserialization;
To facilitate this, a partition map is implemented which is stored as a
separate file. For now, the data file is called `snapstore.bin` and the
partition metadata file is called partmap.
The partition map (`PartMap`) contains a vector of `Partition` objects.
This object stores 'markers' (`len`) which are ideally byte positions
or offsets that demarcate the locations of the individual namespaces.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Previously we were cloning data before writing to disk which
caused slowdowns and lead to higher memory usage. This is an
attempt to fix this behavior.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Also the deps were upgraded (there's no point of dependabot creating
multiple commits for upgrading deps)
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
We previously made a big mistake: we tried to set poisoned to true, an
operation that requires a write lock, without dropping the read lock.
This commit ensures that we first drop `rlock` before doing anything
else. At the same time, if BGSAVE has failed, we'll just shut down the
service.
The error descriptions were improved for a failed `get_saved`
operation. And finally, a couple of variables in `MKSNAP` were renamed
to make them sound more sensible.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
For this, `Writable` was implemented for `&'static [u8]`.
Although this won't have a very noticeable impact on performance, we
will stick to using references instead of cloning the data and then
referencing it again.
Along with this, the docs for MKSNAP were updated.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Let's say that BGSAVE fails for some reason or the other. As it is
our responsibility to ensure the integrity of the user's data and
to enforce the reliability of the overall database, we should
immediately stop accepting writes, so that new changes aren't made
since the last time flushing data to the disk succeeded.
The functions under kvengine were modified as a consequence, and they
will return an "Internal Server Error" which is response code 5 if
the database is poisoned.
To put it in one line: If running BGSAVE fails, we'll `poison` the
in-memory table and stop accepting write operations, which ideally is
any operation that attempts to obtain a write lock.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
There is no merit in having a redundant `enabled` key under
`snapshot` in the configuration file; if a `snapshot` key is present,
it is inherently obvious that snapshotting is enabled
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Instead of checking to see if the remote snapshots directory exists
whenever MKSNAP is called, we'll create the directory when the server
starts up
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Since creating snapshots is quite an important utility,
there may be scenarios where creating one may be needed,
even if it is disabled on the server side. This commit
enables such snapshots to be created. This is achieved by
enabling MKSNAP to accept two arguments, where a 'named'
snapshot can be created, which is our "special" snapshot.
All these "special" snapshots are stored in a separate
"snapshots/remote" dir that is ignored by the
`SnapshotEngine`.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Following on from ab9561258e, we'll allow
the user to configure snapshots and BGSAVE via command-line arguments
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Until now, the database server could only be configured via the
configuration file. This commit enables the host, port and noart
options to be configured via command-line arguments.
This is important as there may be scenarios where creating a file
presents a challenge to the user.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
The user can now run `tdb -r <snapshotname>` to restore data from the
snapshot. Also, we'll show a note in the logs when trying to restore from
a snapshot
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
This module isn't doing anything good staying here, as it does nothing!
More work needs to be done on the file storage format internally before
a public rollout.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
Upstream changes in tokio have required several changes.
For example, `delay_until` was renamed to `sleep_until`. Similarly,
`notify` was renamed to `notify_one`.
Also, in tdb-macros, the `runtime::Builder::new()` line was changed into
`runtime::Builder::new_multi_thread()` due to changes upstream
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
If the duration for a periodic operation is set to zero in the
config file then it is likely that the thread would crash.
We want to prevent this at all costs, which is why
we're adding this check.
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
We don't need tests for MKSNAP when it is enabled as we already have
tests for snapshotting in `diskstore::snapshot`
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
To streamline the actions we'll make sure that mksnap returns code 3
if an incorrect number of arguments are provided
The actiondoc for the `mksnap` action was added and the contributing
guide was also updated
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
In `cli` other errors are now formatted in a `[ERR]` format
Also the documentation across the project was updated
Signed-off-by: Sayan Nandan <nandansayan@outlook.com>
`CoreDB` now has a `SnapshotStatus` object for this purpose
This object holds the current state of the service in an `RWLock`
When `mksnap` is called, `SnapEngine` sets `in_progress` to true
When the snapshot is created, `SnapEngine` sets `in_progress` to false
This prevents multiple entities from creating snapshots at the same time
The proc_macro can now be applied on modules only
All functions within the module will be considered to be a test
This has the advantage of not having to flag every test function
There can be use cases where the user wants to keep all the snapshots
This commit adds a way to keep all the snapshots, by setting maxtop to 0
When `maxtop` is set to 0, all the snapshots will be kept