13782b8ba8 docs: add perf section to developer docs (James O'Beirne)
58180b5fd4 tests: add utility to easily profile node performance with perf (James O'Beirne)
Pull request description:
Adds a context manager to easily (and selectively) profile node performance during functional test execution using `perf`.
While writing some tests, I encountered some odd bitcoind slowness. I wrote up a utility (`TestNode.profile_with_perf`) that generates performance diagnostics for a node by running `perf` during the execution of a particular region of test code.
`perf` usage is detailed in the excellent (and sadly unmerged) https://github.com/bitcoin/bitcoin/pull/12649; all due props to @eklitzke.
### Example
```python
with node.profile_with_perf("large-msgs"):
for i in range(200):
node.p2p.send_message(some_large_msg)
node.p2p.sync_with_ping()
```
This generates a perf data file in the test node's datadir (`/tmp/testtxmpod0y/node0/node-0-TestName-large-msgs.perf.data`).
Running `perf report` generates nice output about where the node spent most of its time while running that part of the test:
```bash
$ perf report -i /tmp/testtxmpod0y/node0/node-0-TestName-large-msgs.perf.data --stdio \
| c++filt \
| less
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 135 of event 'cycles:pp'
# Event count (approx.): 1458205679493582
#
# Children Self Command Shared Object Symbol
# ........ ........ ............... ................... ........................................................................................................................................................................................................................................................................
#
70.14% 0.00% bitcoin-net bitcoind [.] CNode::ReceiveMsgBytes(char const*, unsigned int, bool&)
|
---CNode::ReceiveMsgBytes(char const*, unsigned int, bool&)
70.14% 0.00% bitcoin-net bitcoind [.] CNetMessage::readData(char const*, unsigned int)
|
---CNetMessage::readData(char const*, unsigned int)
CNode::ReceiveMsgBytes(char const*, unsigned int, bool&)
35.52% 0.00% bitcoin-net bitcoind [.] std::vector<char, zero_after_free_allocator<char> >::_M_fill_insert(__gnu_cxx::__normal_iterator<char*, std::vector<char, zero_after_free_allocator<char> > >, unsigned long, char const&)
|
---std::vector<char, zero_after_free_allocator<char> >::_M_fill_insert(__gnu_cxx::__normal_iterator<char*, std::vector<char, zero_after_free_allocator<char> > >, unsigned long, char const&)
CNetMessage::readData(char const*, unsigned int)
CNode::ReceiveMsgBytes(char const*, unsigned int, bool&)
...
```
Tree-SHA512: 9ac4ceaa88818d5eca00994e8e3c8ad42ae019550d6583972a0a4f7b0c4f61032e3d0c476b4ae58756bc5eb8f8015a19a7fc26c095bd588f31d49a37ed0c6b3e
When running tests with --usecli, unify the conversion from argument objects to
strings using a new function arg_to_cli(). This fixes boolean arguments when
using named arguments.
Also use json.dumps() to get the string values for arguments that are dicts and
lists so that bitcoind's JSON parser does not become confused.
Introduces `TestNode.profile_with_perf()` context manager which
samples node execution to produce profiling data.
Also introduces a test framework flag, `--perf`, which will run
perf on all nodes for the duration of a given test.
28479f926f qa: Test bitcond shutdown (João Barbosa)
8d3f46ec39 http: Remove timeout to exit event loop (João Barbosa)
e98a9eede2 http: Remove unnecessary event_base_loopexit call (João Barbosa)
6b13580f4e http: Unlisten sockets after all workers quit (João Barbosa)
18e9685816 http: Send "Connection: close" header if shutdown is requested (João Barbosa)
02e1e4eff6 rpc: Add wait argument to stop (João Barbosa)
Pull request description:
Fixes#11777. Reverts #11006. Replaces #13501.
With this change the HTTP server will exit gracefully, meaning that all requests will finish processing and sending the response, even if this means to wait more than 2 seconds (current time allowed to exit the event loop).
Another small change is that connections are accepted even when the server is stopping, but HTTP requests are rejected. This can be improved later, especially if chunked replies are implemented.
Briefly, before this PR, this is the order or events when a request arrives (RPC `stop`):
1. `bufferevent_disable(..., EV_READ)`
2. `StartShutdown()`
3. `evhttp_del_accept_socket(...)`
4. `ThreadHTTP` terminates (event loop exits) because there are no active or pending events thanks to 1. and 3.
5. client doesn't get the response thanks to 4.
This can be verified by applying
```diff
// Event loop will exit after current HTTP requests have been handled, so
// this reply will get back to the client.
StartShutdown();
+ MilliSleep(2000);
return "Bitcoin server stopping";
}
```
and checking the log output:
```
Received a POST request for / from 127.0.0.1:62443
ThreadRPCServer method=stop user=__cookie__
Interrupting HTTP server
** Exited http event loop
Interrupting HTTP RPC server
Interrupting RPC
tor: Thread interrupt
Shutdown: In progress...
torcontrol thread exit
Stopping HTTP RPC server
addcon thread exit
opencon thread exit
Unregistering HTTP handler for / (exactmatch 1)
Unregistering HTTP handler for /wallet/ (exactmatch 0)
Stopping RPC
RPC stopped.
Stopping HTTP server
Waiting for HTTP worker threads to exit
msghand thread exit
net thread exit
... sleep 2 seconds ...
Waiting for HTTP event thread to exit
Stopped HTTP server
```
For this reason point 3. is moved right after all HTTP workers quit. In that moment HTTP replies are queued in the event loop which keeps spinning util all connections are closed. In order to trigger the server side close with keep alive connections (implicit in HTTP/1.1) the header `Connection: close` is sent if shutdown was requested. This can be tested by
```
bitcoind -regtest
nc localhost 18443
POST / HTTP/1.1
Authorization: Basic ...
Content-Type: application/json
Content-Length: 44
{"jsonrpc": "2.0","method":"stop","id":123}
```
Summing up, this PR:
- removes explicit event loop exit — event loop exits once there are no active or pending events
- changes the moment the listening sockets are removed — explained above
- sends header `Connection: close` on active requests when shutdown was requested which is relevant when it's a persistent connection (default in HTTP 1.1) — libevent is aware of this header and closes the connection gracefully
- removes event loop explicit break after 2 seconds timeout
Tree-SHA512: 4dac1e86abe388697c1e2dedbf31fb36a394cfafe5e64eadbf6ed01d829542785a8c3b91d1ab680d3f03f912d14fc87176428041141441d25dcb6c98a1e069d8
d20a9fa13d tests: add tests for invalid P2P messages (James O'Beirne)
62f94d39f8 tests: add P2PConnection.send_raw_message (James O'Beirne)
5aa31f6ef2 tests: add utility to assert node memory usage hasn't increased (James O'Beirne)
Pull request description:
- Adds `p2p_invalid_messages.py`: tests based on behavior for dealing with invalid and malformed P2P messages. Includes a test verifying that we can't DoS a node by spamming it with large invalid messages.
- Adds `TestNode.assert_memory_usage_stable`: a context manager that allows us to ensure memory usage doesn't significantly increase on a node during some test.
- Adds `P2PConnection.send_raw_message`: which allows us to construct and send messages with tweaked headers.
Tree-SHA512: 720a4894c1e6d8f1551b2ae710e5b06c9e4f281524623957cb01599be9afea82671dc26d6152281de0acb87720f0c53b61e2b27d40434d30e525dd9e31fa671f
Adds a utility to get resident set size memory usage for a test
node and a context manager that allows assertions based upon
maximum memory use increase.
This requires a small changes to a few tests, but means that
deterministic addresses will always be imported (unless setup_nodes
behaviour is explicitly overridden).
Adds a generate() method to the TestNode class in the test framework.
This method intercepts calls to generate, imports a dewterministic
private key to the node and then calls generatetoaddress to generate the
block to that address.
Note that repeated calls to importprivkey for the same private keys are
no-ops, so it's fine to call the generate() method many times.
c1dde3a949 No longer shutdown after encrypting the wallet (Andrew Chow)
d7637c5a3f After encrypting the wallet, reload the database environment (Andrew Chow)
5d296ac810 Add function to close all Db's and reload the databae environment (Andrew Chow)
a769461d5e Move BerkeleyEnvironment deletion from internal method to callsite (Andrew Chow)
Pull request description:
This is the replacement for #11678 which implements @ryanofsky's [suggestion](https://github.com/bitcoin/bitcoin/pull/11678#pullrequestreview-76464511).
Shutting down the software was to prevent the BDB environment from writing unencrypted private keys to disk in the database log files, as was noted [here](https://bitcointalk.org/index.php?topic=51474.msg616068#msg616068). This PR replaces the shutdown behavior with a CDBEnv flush, close, and reopen which achieves the same effect: everything is cleanly flushed and closed, the log files are removed, and then the environment reopened to continue normal operation.
To ensure that no unencrypted private keys are in the log files after encrypting the wallet, I wrote [this script](https://gist.github.com/achow101/7f7143e6c3d3fdc034d3470e72823e9d) to pull private keys from the original wallet file and searches for these keys in the log files (note that you will have to change your file paths to make it work on your own machine).
As for concerns about private keys being written to slack space or being kept in memory, these behaviors no longer exist after the original wallet encryption PR and the shutting down solution from 2011.
cc @ryanofsky
Tree-SHA512: 34b894283b0677a873d06dee46dff8424dec85a2973009ac9b84bcf3d22d05f227c494168c395219d9aee3178e420cf70d4b3eeacc9785aa86b6015d25758e75
fafe73a626 qa: Raise feature_help timeout to 5s (MarcoFalke)
faabd7bc47 qa: Use files for stdout/stderr to support Windows (MarcoFalke)
facb56ffaf qa: Run gen_rpcauth with sys.executable (MarcoFalke)
fada8966c5 qa: Close stdout and stderr file when node stops (MarcoFalke)
Pull request description:
### qa: Close stdout and stderr file when node stops
Since these files are potentially deleted by the test framework for cleanup, they should be closed first. Otherwise this will lead to errors on Windows when the tests finish successfully.
Side note: After the patch, it is no longer possible to reopen the file on Windows (see https://docs.python.org/3/library/tempfile.html#tempfile.NamedTemporaryFile)
### qa: Run gen_rpcauth with sys.executable
Similar to `test_runner.py`, the `sys.executable` needs to be passed down into subprocesses to pass on native Windows. (Should have no effect on Linux)
### qa: Use files for stdout/stderr to support Windows
It seems that using PIPE is not supported on Windows. Also, it is easier to just use the files that capture the stdout and stderr within the test node class.
Tree-SHA512: ec675012b10705978606b7fcbdb287c39a8e6e3732aae2fa4041d963a3c6993c6eac6a9a3cbd5479514e7d8017fe74c12235d1ed6fed2e8af8f3c71981e91864
fa5587fe71 qa: wait_for_verack by default (MarcoFalke)
Pull request description:
This removes the need to do so manually every time a connection is added.
Tree-SHA512: a46c92cb4df41e30778b42b9fd3dcbd8d2d82aa7503d1213cb1c1165034f648d8caee01c292e2d87d05b0f71696996eef5be8a753f35ab49e5f66b0e3bf29f21
3fc20632a3 qt: Set BLOCK_CHAIN_SIZE = 220 (DrahtBot)
2b6a2f4a28 Regenerate manpages (DrahtBot)
eb7daf4d60 Update copyright headers to 2018 (DrahtBot)
Pull request description:
Some trivial maintenance to avoid having to do it again after the 0.17 branch off.
(The scripts to do this are in `./contrib/`)
Tree-SHA512: 16b2af45e0351b1c691c5311d48025dc6828079e98c2aa2e600dc5910ee8aa01858ca6c356538150dc46fe14c8819ed8ec8e4ec9a0f682b9950dd41bc50518fa
fa67505e1e qa: Quote wallet name for rpc path (MarcoFalke)
Pull request description:
When using external multiwallets they are specified by their full path which might contain non-ascii characters (e.g. umlauts or emojis).
Fix this by url-quoting the path.
Tree-SHA512: 7cc66514579d9f602f88a6817c5ab43a44c6d3711df452dc904173f0bc34e2c0b2c34877497f06b61f6720c532fa183053f54622dc454e316c89cee7eaa72463
By default, libc will print fatal errors to /dev/tty instead of stderr.
Adding the LIBC_FATAL_STDERR_ to the environment variables allows
us to catch libc errors in stderr and test for them.
fa811b0 qa: Normalize executable location (MarcoFalke)
Pull request description:
This removes the need to override the executable locations by just reading them from the config file. Beside making the code easier to read, running individual test on Windows is now possible by default (without providing further command line arguments).
Note: Of course, it is still possible to manually specify the location through the `BITCOIND` environment variable, e.g. `bitcoin-qt`.
Tree-SHA512: bee6d22246796242d747120ca18aaab089f73067de213c9111182561985c5912228a0b0f7f9eec025ecfdb44db031f15652f30d67c489d481c995bb3232a7ac7
75d0e4c544 [qa] Delete cookie file before starting node (Suhas Daftuar)
Pull request description:
When a node is restarted during a test after an unclean shutdown (such
as with -dbcrashratio), it's possible an old cookie file was left
behind. This can cause a race condition when restarting the node, where
the test framework might try to connect using credentials from the
old cookie file, just as the node will generate new credentials and
overwrite the old file.
Delete any such cookie file if present prior to startup.
Tree-SHA512: ae1e8bf8fd20e07c32b0715025693bb28b0e3dd34f328cae4346abf579b0c97b5db1c02782e1c46b7a3b6058d268b6d46b668e847658a6eed0be857ffb0d65dc
If a cookie file exists in a datadir prior to node startup, it must have
been leftover from a prior unclean shutdown. As bitcoind will overwrite
it anyway, delete it before starting up to prevent the test framework
from inadvertently trying to connect using stale credentials.
e36a0c0 [qa] Ensure bitcoind processes are cleaned up when tests end (Suhas Daftuar)
Pull request description:
When tests fail (such as due to a bug in the test, race condition, etc), it's possible that we could follow code paths that bypass our normal node shutdown that occurs in `TestNode.stop_node`. Add a destructor to `TestNode` that cleans this up.
Tree-SHA512: 72e04bc21462ebd0cb346fd1fe0540da454acfbad41923a0b06ea2317e9045b68e58f9adb02d8200891aca89a9d03a022eb72282aeb31a3b3afe7c6843a4b450
c8330d4 qa: Use node.datadir instead of tmpdir in test framework (MarcoFalke)
Pull request description:
Commit c53c9831ee introduced the utility function `get_datadir_path`, however not all places in the code use this util function. Using the util function everywhere makes it easier to review pull requests related to the datadir.
This commit replaces datadir path creation with the `datadir` member of `TestNode`, which itself uses `get_datadir_path`.
Tree-SHA512: c75707ab7149d732a6d56152a5813138a33459d3d07577b60b89f2a207c83b7663fac5f203593677c9892d1c23a5eba4bd45c5c4ababf040d720b437240fcddf
fae1374 qa: Allow for partial_match when checking init error (MarcoFalke)
5812273 [Tests] Require exact match in assert_start_raises_init_eror() (John Newbery)
0ec08a6 [Tests] Move assert_start_raises_init_error method to TestNode (John Newbery)
Pull request description:
Extracted from #12379, because the changes are important on their own.
This allows for exact testing, since the match can be specified with a strict regex. Internal details (such as exact formatting of the error message) can still be fuzzed away by regex wildcards.
Tree-SHA512: 605d2c9c42362a32d42321b066637577a026d0bb8cfc1c9f5737a4ca6503ffe85457a5122cea6e1101053ccc6c8aa1bbae3602e1fa7d2988bf7d5c275f412f66
Prevents OSX firewall allow-this-application-to-accept-inbound-connections
permission popups and is generally safer.
To prevent binding to 127.0.0.1, set self.bind_to_localhost_only = False.
fae7b14a04 qa: Make TestNodeCLI command optional in send_cli (MarcoFalke)
ffffb10a9f qa: Rename cli.args to cli.options (MarcoFalke)
Pull request description:
Makes the `command` optional, since there are valid bitcoin-cli calls that have no `command`:
* `bitcoin-cli -?`
* `bitcoin-cli -getinfo`
* ...
Also, rename self.args to self.options, since that is the name in the `bitcoin-cli -help` documentation.
Tree-SHA512: f49c06024e78423301d70782946d47c0fb97a26876afba0a1f71ed329f5d7124aee4c2df520c7af74079bf9937851902f7be9c54abecc28dc29274584804d46c
Also warn if bitcoind is configured to use a relative -datadir path.
Specifying paths relative to the current working directory in a daemon process
can be dangerous, because files can fail to be located even if the
configuration doesn't change, but the daemon is started up differently.
Specifying a relative -datadir now adds a warning to the debug log. It would
not be backwards-compatible to forbid relative -datadir paths entirely, and it
could also be also inconvenient for command line testing.
Specifying a relative -walletdir now results in a startup error. But since the
-walletdir option is new in 0.16.0, there should be no compatibility issues.
Another reason not to use working directory paths for -walletdir specifically
is that the default -walletdir is a "wallets" subdirectory inside the datadir,
so it could be surprising that setting -walletdir manually would choose a
directory rooted in a completely different location.
test_framework accepts a new --usecli parameter. Running the test with
this parameter will cause all RPCs to be sent through bitcoin-cli rather
than directly over http. By default, individual test cases do not
support --usecli, and self.supports_cli must be set to True in the
set_test_params method.
We can make supports_cli default to True in future once we know which
tests will fail with use_cli.
Change TestNodeCLI.__call__() to return a new instance instead of modifying the
existing instance. This way, it's possible to create different cli objects that
have their own options (for example -rpcwallet options to connect to different
wallets), and options set for a single call (`node.cli(options).method(args)`)
will no longer leak into future calls.
This makes NodeConnCB a subclass of NodeConn, and
removes the need for the client code to know
anything about the implementation details of NodeConnCB.
NodeConn can now be swapped out for any other implementation
of a low-level connection without changing client code.
fafdad0d4 qa: Remove unused NodeConn members (MarcoFalke)
Pull request description:
* `ver_send` and `ver_recv` were completely unused
* `rpc` was only used once, in p2p-segwit. Imo better only pass it to the constructor in that single test
Tree-SHA512: 7f85554d6d0fd2096516ca3c608811d5370da66cde35d8031bdc921607a7a4efdb26355896012f75f713f8df09e28d46ba46be69fd96a5898fabb1a25cbcb8ad
p2p connections can now be added to TestNode instances.
This commit also updates the example test to use the new
p2p interface in TestNode to demonstrate usage.
A future commit will update the existing tests to use p2p through the
TestNode.
7148b74dc [tests] Functional tests must explicitly set num_nodes (John Newbery)
5448a1471 [tests] don't override __init__() in individual tests (John Newbery)
6cf094a02 [tests] Avoid passing around member variables in test_framework (John Newbery)
36b626867 [tests] TestNode: separate add_node from start_node (John Newbery)
be2a2ab6a [tests] fix - use rpc_timeout as rpc timeout (John Newbery)
Pull request description:
Some additional tidyups after the introduction of TestNode:
- commit 1 makes TestNode use the correct rpc timeout. This should have been included in #11077
- commit 2 separates `add_node()` from `start_node()` as originally discussed here: https://github.com/bitcoin/bitcoin/pull/10556#discussion_r121161453 with @kallewoof . The test writer no longer needs to assign to `self.nodes` when starting/stopping nodes.
- commit 3 adds a `set_test_params()` method, so individual tests don't need to override `__init__()` and call `super().__init__()`
Tree-SHA512: 0adb030623b96675b5c29e2890ce99ccd837ed05f721d0c91b35378c5ac01b6658174aac12f1f77402e1d38b61f39b3c43b4df85c96952565dde1cda05b0db84
Separates the act of creating a TestNode object from starting the node.
The test_framework now keeps track of its list of TestNodes, and test
writers can call start_node() and stop_node() without having to update
the self.nodes list.
c6ec4358a [tests] Add bitcoin_cli.py test script (John Newbery)
b23549f6e [tests] add TestNodeCLI class for calling bitcoin-cli for a node (John Newbery)
Pull request description:
We don't test bitcoin-cli at all. That means that we can miss inconsistencies between the bitcoin-cli client and the RPC interface, such as #10698 and #10747. It also means that the various bitcoin-cli options and features are untested and regressions could be silently introduced.
Let's fix that.
This PR adds bitcoin-cli testing in the python functional test_framework:
1. Add a bitcoin_cli.py test script that tests bitcoin-cli. At the moment it only tests that the result of `getinfo` is the same if you run it as an RPC or through bitcoin-cli, but can easily be extended to test additional bitcoin-cli features
**EDIT: `--usecli` option is moved to a separate PR. This PR now only covers the bitcoin_cli.py test.**
2. ~Add a `--usecli` option to the test framework. This changes the test to use bitcoin-cli for all RPC calls instead of using direct HTTP requests. This is somewhat experimental. It works for most tests, but there are some cases where it can't work transparently because:~
- ~the testcase is asserting on a specific error code, and bitcoin-cli returns a different error code from the direct RPC~
- ~we're sending a very large RPC request (eg `submitblock`) and it can't be serialized into a shell bitcoin-cli call.~
~I think that even though `--usecli` doesn't work on all tests, it's still a useful experimental feature. Future potential enhancements:~
- ~enhance the framework to automatically skip tests that are known to fail with bitcoin-cli if the `--usecli` option is used.~
- ~run a subset of tests in Travis with `-usecli`~
This builds on and requires the `TestNode` PR #10711 . As an aside, this is a good demonstration of how tidy it is to add additional features/interfaces now that test node logic/state is encapsulated in a TestNode class.
Addresses #10791
Tree-SHA512: a1e6be12e8e007f6f67b3d3bbcd142d835787300831eb38e6027a1ad25ca9d79c4bc99a41b19e31ee95205cba1b3b2d21a688b5909316aad70bfc2b4eb6d8a52
Fixes a couple of bugs from the introduction of TestNode:
- test scripts were no longer able to specify a custom timeout for
starting a node. Therefore tests with nodes that take a long time to
start up (eg pruning.py) would fail.
- the test for whether a node has failed on start up was broken
by changing 'assert x is None' to 'assert not x'. Since
subprocess.poll() can return None (indicating the node is still running)
or 0 (indicating the node exited with return code 0), this was a
regression.
When running the tests locally with a parallelism of 4 on an otherwise
busy system, RPC can take quite a wait to come up.
Change the timeout to 60 seconds just to be safe.
TestNode is a class responsible for all state related to a bitcoind node
under test. It stores local state, is responsible for tracking the
bitcoind process and delegates unrecognised messages to the RPC
connection.
This commit changes start_nodes and stop_nodes to start and stop the
bitcoind nodes in parallel, making test setup and teardown much faster.
