Caching the base58check encoded version of an `AccountID` has
performance advantages, because because of the computationally
heavy cost associated with the conversion, which requires the
application of SHA-256 twice.
This commit makes the cache significantly more efficient in terms
of memory used: it eliminates the map, using a vector with a size
that is determined by the configured size of the node, and a hash
function to directly map any given `AccountID` to a specific slot
in the cache; the eviction policy is simple: in case of collision
the existing entry is removed and replaced with the new data.
Previously, use of the cache was optional and required additional
effort by the programmer. Now the cache is automatic and does not
require any additional work or information.
The new cache also utilizes a 64-way spinlock, to help reduce any
contention that the pressure on the cache would impose.
- Only duplicate records from archive to writable during online_delete.
- Log duration of nodestore reads.
- Include nodestore counters in perf_log output.
- Remove gratuitous nodestore activity counting.
- Report initial sync duration in server_info and perfLog.
- Report state_accounting in perfLog.
- Make state_accounting durations more accurate.
- Parallel ledger loader.
- Config parameter to load ledgers on start.
The pathfinding engine requires pre-building large tables which is a
resource-intensive operation. Typically, one would not expect that a
server configured as a validator would also support pathfinding APIs
and so, building those tables by default wastes resources.
This commit, if merged, will disable pathfinding on servers that are
configured as validators, unless the server operator opts to support
it explicitly, by configuring the `[path_search_max]` parameter.
Validator operators that wish to support pathfinding on a validator
and want to use the default values can add the following stanza to
their server's configuration file:
[path_search_max]
7
The existing calculation would limit the maximum number of threads
that would be created by default to at most 6; this may have been
reasonable a few years ago, but given both the load on the network
as of today and the increase in the number of CPU cores, the value
should be revisited.
This commit, if merged, changes the default calculation for nodes
that are configured as `large` or `huge` to allow for up to twelve
threads.
The "sweep interval" is the amount of time between successive sweeps of
of various in-memory data structures to remove stale items.
Prior to this commit, the interval was automatically adjusted, based on
the value of the `[node_size]` option in a server's configuration file.
If merged, this commit introduces a new configuration option that makes
it possible for a server operator to adjust the sweep interval and make
a CPU/memory tradeoff:
[sweep_interval]
<integer>
The specified value represents the number of seconds between successive
sweeps. The range of valid values is between 10 and 600.
Important operator notes:
This is an advanced configuration option that should not be used unless
there is empirical data which suggests that the default sweep frequency
is either resulting in performance problems or is causing undue load to
the server.
Note that adjusting the sweep interval may not have the intended effect
on the server. Lower values will not always translate to a reduction of
memory usage and higher values will not always translate to a reduction
of CPU usage and/or load.
The existing logic involves every server sending every transaction
that it receives to all its peers (except the one that it received
a transaction from).
This commit instead uses a randomized algorithm, where a node will
randomly select peers to relay a given transaction to, caching the
list of transaction hashes that are not relayed and forwading them
to peers once every second. Peers can then determine whether there
are transactions that they have not seen and can request them from
the node which has them.
It is expected that this feature will further reduce the bandwidth
needed to operate a server.
The following changes were made:
- Removed dependency on template defined in beast detail namespace.
- Removed Section::find() method which had an obsolete interface.
- Made Section::get<>() easier to use for the common case of
retrieving a std::string. The revised get() method replaces old
calls to Section::find().
- Provided a default template parameter to free function
get<>(Section config, std::string name) so it stays similar to
Section::get<>().
Then the rest of the code was adapted to these changes.
- Calls to Section::find() were replaced with calls to Section::get.
- Unnecessary get<std::string>() arguments were reduced to get().
These changes dug up an interesting artifact in the SHAMap unit
tests. I'm not sure why the tests were working before, but there
was a problem with the case of a Section key. The unit test is
fixed.
The `[node_size]` configuration parameter is used to tune various
parameters based on the hardware that the code is running on. The
parameter can take five distinct values: `tiny`, `small`, `medium`,
`large` and `huge`.
The default value in the code is `tiny` but the default configuration
file sets the value to `medium`. This commit attempts to detect the
amount of RAM on the system and adjusts the node size default value
based on the amount of RAM and the number of hardware execution
threads on the system.
The decision matrix currently used is:
| | 1 | 2 or 3 | ≥ 4 |
|:-------:|:----:|:------:|:------:|
| > ~8GB | tiny | tiny | tiny |
| > ~12GB | tiny | small | small |
| > ~16GB | tiny | small | medium |
| > ~24GB | tiny | small | large |
| > ~32GB | tiny | small | huge |
Some systems exclude memory reserved by the the hardware, the kernel
or the underlying hypervisor so the automatic detection code may end
up determining the node_size to be one less than "appropriate" given
the above table.
The detection algorithm is simplistic and does not take into account
other relevant factors. Therefore, for production-quality servers it
is recommended that server operators examine the system holistically
and determine what the appropriate size is instead of relying on the
automatic detection code.
To aid server operators, the node size will now be reported in the
`server_info` API as `node_size` when the command is invoked in
'admin' mode.
* Add a new operating mode to rippled called reporting mode
* Add ETL mechanism for a reporting node to extract data from a p2p node
* Add new gRPC methods to faciliate ETL
* Use Postgres in place of SQLite in reporting mode
* Add Cassandra as a nodestore option
* Update logic of RPC handlers when running in reporting mode
* Add ability to forward RPCs to a p2p node
- Add validation/proposal reduce-relay feature negotiation to
the handshake
- Make squelch duration proportional to a number of peers that
can be squelched
- Refactor makeRequest()/makeResponse() to facilitate handshake
unit-testing
- Fix compression enable flag for inbound peer
- Fix compression algorithm parsing in the header parser
- Fix squelch duration in onMessage(TMSquelch)
This commit fixes 3624, fixes 3639 and fixes 3641
This comment explains this patch and the associated patches
that should be folded into it. This paragraph should be removed
when the patches are folded after review.
This change significantly improves ledger sync and fetch
times while reducing memory consumption. The change affects
the code from that begins with SHAMap::getMissingNodes and runs
through to Database::threadEntry.
The existing code issues a number of async fetches which are then
handed off to the Database's pool of read threads to execute.
The results of each read are placed in the Database's positive
and negative caches. The caller waits for all reads to complete
and then retrieves the results out of these caches.
Among other issues, this means that the results of the first read
cannot be processed until the last read completes. Additionally,
all the results must sit in memory.
This patch changes the behavior so that each read operation has a
completion handler associated with it. The completion of the read
calls the handler, allowing the results of each read to be
processed as it completes. As this was the only reason the
negative and positive caches were needed, they can now be removed.
The read generation code is also no longer needed and is removed.
The batch fetch logic was never implemented or supported and is
removed.
When evaluating the fitness and usefulness of an outbound peer, the code
would incorrectly calculate the amount of time that the peer spent in
a non-useful state.
This commit, if merged, corrects the calculation and makes the timeout
values configurable by server operators.
Two new options are introduced in the 'overlay' stanza of the config
file. The default values, in seconds, are:
[overlay]
max_unknown_time = 600
max_diverged_time = 300
This commit replaces the `peers_max` configuration element which had
a predetermined split between incoming and outgoing connections with
two new configuration options, `peers_in_max` and `peers_out_max`,
which server operators can use to explicitly control the number of
incoming and outgoing peer slots.
This commit introduces a new configuration option that server
operators can set. The value is communicated to other servers
and is also reported via the `server_info` API.
The value is meant to allow third-party applications or tools
to group servers together. For example, a tool that visualizes
the network's topology can group servers together.
Similar to the "Domain" field in validator manifests, an operator
can claim any domain. Prior to relying on the value returned, the
domain should be verified by retrieving the xrp-ledger.toml file
from the domain and looking for the server's public key in the
`nodes` array.
The job queue can impose limits of how many jobs of a particular
type can be queued.
This commit makes the previously hard-coded limit associated with
transactions configurable by the server's operator. Servers that
have increased memory capacity or which expect to see an influx
of transactions can increase the number of transactions their
server will be able to queue.
This commit fixes#3556.
With few exceptions, servers will typically receive multiple copies
of any given message from its directly connected peers. For servers
with several peers this can impact the processing latency and force
it to do redundant work. Proposal and validation messages are often
relayed with extremely high redundancy.
This commit, if merged, introduces experimental code that attempts
to optimize the relaying of proposals and validations by allowing
servers to instruct their peers to "squelch" delivery of selected
proposals and validations. Servers making squelching decisions by
a process that evaluates the fitness and performance of a given
server and randomly selecting a subset of the best candidates.
The experimental code is presently disabled and must be explicitly
enabled by server operators that wish to test it.
* Document delete_batch, back_off_milliseconds, age_threshold_seconds.
* Convert those time values to chrono types.
* Fix bug that ignored age_threshold_seconds.
* Add a "recovery buffer" to the config that gives the node a chance to
recover before aborting online delete.
* Add begin/end log messages around the SQL queries.
* Add a new configuration section: [sqlite] to allow tuning the sqlite
database operations. Ignored on full/large history servers.
* Update documentation of [node_db] and [sqlite] in the
rippled-example.cfg file.
Resolves#3321
* The amendment ballot counting code contained a minor technical
flaw, caused by the use of integer arithmetic and rounding
semantics, that could allow amendments to reach majority with
slightly less than 80% support. This commit introduces an
amendment which, if enabled, will ensure that activation
requires at least 80% support.
* This commit also introduces a configuration option to adjust
the amendment activation hysteresis. This option is useful on
test networks, but should not be used on the main network as
is a network-wide consensus parameter that should not be
changed on a per-server basis; doing so can result in a
hard-fork.
Fixes#3396
In deciding whether to relay a proposal or validation, a server would
consider whether it was issued by a validator on that server's UNL.
While both trusted proposals and validations were always relayed,
the code prioritized relaying of untrusted proposals over untrusted
validations. While not technically incorrect, validations are
generally more "valuable" because they are required during the
consensus process, whereas proposals are not, strictly, required.
The commit introduces two new configuration options, allowing server
operators to fine-tune the relaying behavior:
The `[relay_proposals]` option controls the relaying behavior for
proposals received by this server. It has two settings: "trusted"
and "all" and the default is "trusted".
The `[relay_validations]` options controls the relaying behavior for
validations received by this server. It has two settings: "trusted"
and "all" and the default is "all".
This change does not require an amendment as it does not affect
transaction processing.
* Peers negotiate compression via HTTP Header "X-Offer-Compression: lz4"
* Messages greater than 70 bytes and protocol type messages MANIFESTS,
ENDPOINTS, TRANSACTION, GET_LEDGER, LEDGER_DATA, GET_OBJECT,
and VALIDATORLIST are compressed
* If the compressed message is larger than the uncompressed message
then the uncompressed message is sent
* Compression flag and the compression algorithm type are included
in the message header
* Only LZ4 block compression is currently supported
Treat all `#` characters in config files as comments (and remove)
*unless* the `#` is immediately preceded by `\`. Write a warning
to log file when trailing comments are found/ignored in the config
to let operators know that the treatment of trailing `#` has changed.
Fixes#3121
The `node_size` configuration option is used to automatically
configure various parameters (cache sizes, timeouts, etc) for
the server.
A previous commit included changes that caused incorrect values
to be returned which can result in sub-optimal performance that
can manifest as difficulty syncing to the network, or increased
disk I/O and/or memory usage. The problem was introduced with
commit 66fad62e66.
This commit, if merged, fixes the code to ensure that the correct
values are returned and introduces a compile-time check to prevent
this issue from reoccurring.
* replace boost::beast::detail::iequals with boost::iequals
* replace deprecated `buffers` function with `make_printable`
* replace boost::beast::detail::ascii_tolower with lambda
* add missing includes
If the number of peers a server has is below the configured
minimum peer limit, this commit will properly transition the
server's state to "disconnected".
The default limit for the minimum number of peers required was
0 meaning that a server that was connected but lost all its
peers would never transition to disconnected, since it could
never drop below zero peers.
This commit redefines the default minimum number of peers to 1
and produces a warning if the server is configured in a way
that will prevent it from ever achieving sufficient connectivity.
* Adds local file:// URL support to the [validator_list_sites] stanza.
The file:// URL must not contain a hostname. Allows a rippled node
operator to "sideload" a new list if their node is unable to reach
a validator list's web site before an old list expires. Lists
loaded from a file will be validated in the same way a downloaded
list is validated.
* Generalize file/dir "guards" from Config test so they can be reused
in other tests.
* Check for error when reading validators.txt. Saves some parsing and
checking of an empty string, and will give a more meaningful error.
* Completes RIPD-1674.
In order to facilitate transaction signing, `rippled` offers the `sign` and
`sign_for` and `submit` commands, which, given a seed, can be used to sign or
sign-and-submit transactions. These commands are accessible from the command
line, as well as over the WebSocket and RPC interfaces that `rippled` can be
configured to provide.
These commands, unfortunately, have significant security implications:
1. They require divulging an account's seed (commonly known as a "secret
key") to the server.
2. When executing these commands against remote servers, the seeds can be
transported over clear-text links.
3. When executing these commands over the command line, the account
seed may be visible using common tools that show running processes
and may potentially be inadvertently stored by system monitoring
tools or facilities designed to maintain a history of previously
typed commands.
While this commit cannot prevent users from issuing these commands to a
server, whether locally or remotely, it restricts the `sign` and `sign_for`
commands, as well as the `submit` command when used to sign-and-submit,
so that they require administrative privileges on the server.
Server operators that want to allow unrestricted signing can do so by
adding the following stanza to their configuration file:
[signing_support]
true
Ripple discourages server operators from doing so and advises against using
these commands, which will be removed in a future release. If you rely on
these commands for signing, please migrate to a standalone signing solution
as soon as possible. One option is to use `ripple-lib`; documentation is
available at https://developers.ripple.com/rippleapi-reference.html#sign.
If the commands are administratively enabled, the server includes a warning
on startup and adds a new field in the resulting JSON, informing the caller
that the commands are deprecated and may become unavailable at any time.
Acknowledgements:
Jesper Wallin for reporting this issue to Ripple.
Bug Bounties and Responsible Disclosures:
We welcome reviews of the rippled code and urge researchers to responsibly
disclose any issues that they may find. For more on Ripple's Bug Bounty
program, please visit: https://ripple.com/bug-bounty
Enable all supported amendments in Env by default. Rename `features()`
to `with_features()` and add `all_features_except()` to support feature
subsets in Env. Refactor internal feature handling based on a bitset.
The job queue can automatically tune the number of threads that
it creates based on the number of processors or processor cores
that are available.
The existing tuning was very conservative, limiting the maximum
number of threads to only 6.
Adjust the new algorithm to allow a larger number of threads and
allow server administrators to override the value in the config
file.
