Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
#!/usr/bin/env python3
|
|
|
|
# Copyright (c) 2022 The Bitcoin Core developers
|
|
|
|
# Distributed under the MIT software license, see the accompanying
|
|
|
|
# file COPYING or http://www.opensource.org/licenses/mit-license.php.
|
|
|
|
|
|
|
|
""" Tests the mempool:* tracepoint API interface.
|
|
|
|
See https://github.com/bitcoin/bitcoin/blob/master/doc/tracing.md#context-mempool
|
|
|
|
"""
|
|
|
|
|
|
|
|
from decimal import Decimal
|
|
|
|
|
|
|
|
# Test will be skipped if we don't have bcc installed
|
|
|
|
try:
|
|
|
|
from bcc import BPF, USDT # type: ignore[import]
|
|
|
|
except ImportError:
|
|
|
|
pass
|
|
|
|
|
|
|
|
from test_framework.blocktools import COINBASE_MATURITY
|
|
|
|
from test_framework.messages import COIN, DEFAULT_MEMPOOL_EXPIRY_HOURS
|
|
|
|
from test_framework.p2p import P2PDataStore
|
|
|
|
from test_framework.test_framework import BitcoinTestFramework
|
|
|
|
from test_framework.util import assert_equal
|
|
|
|
from test_framework.wallet import MiniWallet
|
|
|
|
|
|
|
|
MEMPOOL_TRACEPOINTS_PROGRAM = """
|
|
|
|
# include <uapi/linux/ptrace.h>
|
|
|
|
|
|
|
|
// The longest rejection reason is 118 chars and is generated in case of SCRIPT_ERR_EVAL_FALSE by
|
|
|
|
// strprintf("mandatory-script-verify-flag-failed (%s)", ScriptErrorString(check.GetScriptError()))
|
|
|
|
#define MAX_REJECT_REASON_LENGTH 118
|
|
|
|
// The longest string returned by RemovalReasonToString() is 'sizelimit'
|
|
|
|
#define MAX_REMOVAL_REASON_LENGTH 9
|
|
|
|
#define HASH_LENGTH 32
|
|
|
|
|
|
|
|
struct added_event
|
|
|
|
{
|
|
|
|
u8 hash[HASH_LENGTH];
|
2023-06-12 14:47:19 -04:00
|
|
|
s32 vsize;
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
s64 fee;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct removed_event
|
|
|
|
{
|
|
|
|
u8 hash[HASH_LENGTH];
|
|
|
|
char reason[MAX_REMOVAL_REASON_LENGTH];
|
2023-06-12 14:47:19 -04:00
|
|
|
s32 vsize;
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
s64 fee;
|
|
|
|
u64 entry_time;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct rejected_event
|
|
|
|
{
|
|
|
|
u8 hash[HASH_LENGTH];
|
|
|
|
char reason[MAX_REJECT_REASON_LENGTH];
|
|
|
|
};
|
|
|
|
|
|
|
|
struct replaced_event
|
|
|
|
{
|
|
|
|
u8 replaced_hash[HASH_LENGTH];
|
2023-06-12 14:47:19 -04:00
|
|
|
s32 replaced_vsize;
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
s64 replaced_fee;
|
|
|
|
u64 replaced_entry_time;
|
|
|
|
u8 replacement_hash[HASH_LENGTH];
|
2023-06-12 14:47:19 -04:00
|
|
|
s32 replacement_vsize;
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
s64 replacement_fee;
|
|
|
|
};
|
|
|
|
|
|
|
|
// BPF perf buffer to push the data to user space.
|
|
|
|
BPF_PERF_OUTPUT(added_events);
|
|
|
|
BPF_PERF_OUTPUT(removed_events);
|
|
|
|
BPF_PERF_OUTPUT(rejected_events);
|
|
|
|
BPF_PERF_OUTPUT(replaced_events);
|
|
|
|
|
|
|
|
int trace_added(struct pt_regs *ctx) {
|
|
|
|
struct added_event added = {};
|
|
|
|
|
|
|
|
bpf_usdt_readarg_p(1, ctx, &added.hash, HASH_LENGTH);
|
|
|
|
bpf_usdt_readarg(2, ctx, &added.vsize);
|
|
|
|
bpf_usdt_readarg(3, ctx, &added.fee);
|
|
|
|
|
|
|
|
added_events.perf_submit(ctx, &added, sizeof(added));
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int trace_removed(struct pt_regs *ctx) {
|
|
|
|
struct removed_event removed = {};
|
|
|
|
|
|
|
|
bpf_usdt_readarg_p(1, ctx, &removed.hash, HASH_LENGTH);
|
|
|
|
bpf_usdt_readarg_p(2, ctx, &removed.reason, MAX_REMOVAL_REASON_LENGTH);
|
|
|
|
bpf_usdt_readarg(3, ctx, &removed.vsize);
|
|
|
|
bpf_usdt_readarg(4, ctx, &removed.fee);
|
|
|
|
bpf_usdt_readarg(5, ctx, &removed.entry_time);
|
|
|
|
|
|
|
|
removed_events.perf_submit(ctx, &removed, sizeof(removed));
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int trace_rejected(struct pt_regs *ctx) {
|
|
|
|
struct rejected_event rejected = {};
|
|
|
|
|
|
|
|
bpf_usdt_readarg_p(1, ctx, &rejected.hash, HASH_LENGTH);
|
|
|
|
bpf_usdt_readarg_p(2, ctx, &rejected.reason, MAX_REJECT_REASON_LENGTH);
|
|
|
|
|
|
|
|
rejected_events.perf_submit(ctx, &rejected, sizeof(rejected));
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int trace_replaced(struct pt_regs *ctx) {
|
|
|
|
struct replaced_event replaced = {};
|
|
|
|
|
|
|
|
bpf_usdt_readarg_p(1, ctx, &replaced.replaced_hash, HASH_LENGTH);
|
|
|
|
bpf_usdt_readarg(2, ctx, &replaced.replaced_vsize);
|
|
|
|
bpf_usdt_readarg(3, ctx, &replaced.replaced_fee);
|
|
|
|
bpf_usdt_readarg(4, ctx, &replaced.replaced_entry_time);
|
|
|
|
bpf_usdt_readarg_p(5, ctx, &replaced.replacement_hash, HASH_LENGTH);
|
|
|
|
bpf_usdt_readarg(6, ctx, &replaced.replacement_vsize);
|
|
|
|
bpf_usdt_readarg(7, ctx, &replaced.replacement_fee);
|
|
|
|
|
|
|
|
replaced_events.perf_submit(ctx, &replaced, sizeof(replaced));
|
|
|
|
return 0;
|
|
|
|
}
|
2023-10-09 10:38:37 -03:00
|
|
|
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
|
|
class MempoolTracepointTest(BitcoinTestFramework):
|
|
|
|
def set_test_params(self):
|
|
|
|
self.num_nodes = 1
|
|
|
|
self.setup_clean_chain = True
|
|
|
|
|
|
|
|
def skip_test_if_missing_module(self):
|
|
|
|
self.skip_if_platform_not_linux()
|
|
|
|
self.skip_if_no_bitcoind_tracepoints()
|
|
|
|
self.skip_if_no_python_bcc()
|
|
|
|
self.skip_if_no_bpf_permissions()
|
|
|
|
|
|
|
|
def added_test(self):
|
|
|
|
"""Add a transaction to the mempool and make sure the tracepoint returns
|
|
|
|
the expected txid, vsize, and fee."""
|
|
|
|
|
2023-06-23 07:05:34 -04:00
|
|
|
events = []
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
self.log.info("Hooking into mempool:added tracepoint...")
|
|
|
|
node = self.nodes[0]
|
|
|
|
ctx = USDT(pid=node.process.pid)
|
|
|
|
ctx.enable_probe(probe="mempool:added", fn_name="trace_added")
|
2023-10-09 10:38:37 -03:00
|
|
|
bpf = BPF(text=MEMPOOL_TRACEPOINTS_PROGRAM, usdt_contexts=[ctx], debug=0, cflags=["-Wno-error=implicit-function-declaration"])
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def handle_added_event(_, data, __):
|
2023-06-23 07:05:34 -04:00
|
|
|
events.append(bpf["added_events"].event(data))
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
bpf["added_events"].open_perf_buffer(handle_added_event)
|
|
|
|
|
|
|
|
self.log.info("Sending transaction...")
|
|
|
|
fee = Decimal(31200)
|
|
|
|
tx = self.wallet.send_self_transfer(from_node=node, fee=fee / COIN)
|
|
|
|
|
|
|
|
self.log.info("Polling buffer...")
|
|
|
|
bpf.perf_buffer_poll(timeout=200)
|
|
|
|
|
|
|
|
self.log.info("Cleaning up mempool...")
|
|
|
|
self.generate(node, 1)
|
|
|
|
|
|
|
|
self.log.info("Ensuring mempool:added event was handled successfully...")
|
2023-06-23 07:05:34 -04:00
|
|
|
assert_equal(1, len(events))
|
|
|
|
event = events[0]
|
Manage exceptions in bcc callback functions
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
2023-06-05 10:31:31 -04:00
|
|
|
assert_equal(bytes(event.hash)[::-1].hex(), tx["txid"])
|
|
|
|
assert_equal(event.vsize, tx["tx"].get_vsize())
|
|
|
|
assert_equal(event.fee, fee)
|
|
|
|
|
|
|
|
bpf.cleanup()
|
2023-03-29 07:17:29 -03:00
|
|
|
self.generate(self.wallet, 1)
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def removed_test(self):
|
|
|
|
"""Expire a transaction from the mempool and make sure the tracepoint returns
|
|
|
|
the expected txid, expiry reason, vsize, and fee."""
|
|
|
|
|
2023-06-23 07:05:34 -04:00
|
|
|
events = []
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
self.log.info("Hooking into mempool:removed tracepoint...")
|
|
|
|
node = self.nodes[0]
|
|
|
|
ctx = USDT(pid=node.process.pid)
|
|
|
|
ctx.enable_probe(probe="mempool:removed", fn_name="trace_removed")
|
2023-10-09 10:38:37 -03:00
|
|
|
bpf = BPF(text=MEMPOOL_TRACEPOINTS_PROGRAM, usdt_contexts=[ctx], debug=0, cflags=["-Wno-error=implicit-function-declaration"])
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def handle_removed_event(_, data, __):
|
2023-06-23 07:05:34 -04:00
|
|
|
events.append(bpf["removed_events"].event(data))
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
bpf["removed_events"].open_perf_buffer(handle_removed_event)
|
|
|
|
|
|
|
|
self.log.info("Sending transaction...")
|
|
|
|
fee = Decimal(31200)
|
|
|
|
tx = self.wallet.send_self_transfer(from_node=node, fee=fee / COIN)
|
|
|
|
txid = tx["txid"]
|
|
|
|
|
|
|
|
self.log.info("Fast-forwarding time to mempool expiry...")
|
|
|
|
entry_time = node.getmempoolentry(txid)["time"]
|
|
|
|
expiry_time = entry_time + 60 * 60 * DEFAULT_MEMPOOL_EXPIRY_HOURS + 5
|
|
|
|
node.setmocktime(expiry_time)
|
|
|
|
|
|
|
|
self.log.info("Triggering expiry...")
|
|
|
|
self.wallet.get_utxo(txid=txid)
|
|
|
|
self.wallet.send_self_transfer(from_node=node)
|
|
|
|
|
|
|
|
self.log.info("Polling buffer...")
|
|
|
|
bpf.perf_buffer_poll(timeout=200)
|
|
|
|
|
|
|
|
self.log.info("Ensuring mempool:removed event was handled successfully...")
|
2023-06-23 07:05:34 -04:00
|
|
|
assert_equal(1, len(events))
|
|
|
|
event = events[0]
|
Manage exceptions in bcc callback functions
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
2023-06-05 10:31:31 -04:00
|
|
|
assert_equal(bytes(event.hash)[::-1].hex(), txid)
|
|
|
|
assert_equal(event.reason.decode("UTF-8"), "expiry")
|
|
|
|
assert_equal(event.vsize, tx["tx"].get_vsize())
|
|
|
|
assert_equal(event.fee, fee)
|
|
|
|
assert_equal(event.entry_time, entry_time)
|
|
|
|
|
|
|
|
bpf.cleanup()
|
2023-03-29 07:17:29 -03:00
|
|
|
self.generate(self.wallet, 1)
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def replaced_test(self):
|
|
|
|
"""Replace one and two transactions in the mempool and make sure the tracepoint
|
|
|
|
returns the expected txids, vsizes, and fees."""
|
|
|
|
|
2023-06-23 07:05:34 -04:00
|
|
|
events = []
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
self.log.info("Hooking into mempool:replaced tracepoint...")
|
|
|
|
node = self.nodes[0]
|
|
|
|
ctx = USDT(pid=node.process.pid)
|
|
|
|
ctx.enable_probe(probe="mempool:replaced", fn_name="trace_replaced")
|
2023-10-09 10:38:37 -03:00
|
|
|
bpf = BPF(text=MEMPOOL_TRACEPOINTS_PROGRAM, usdt_contexts=[ctx], debug=0, cflags=["-Wno-error=implicit-function-declaration"])
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def handle_replaced_event(_, data, __):
|
2023-06-23 07:05:34 -04:00
|
|
|
events.append(bpf["replaced_events"].event(data))
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
bpf["replaced_events"].open_perf_buffer(handle_replaced_event)
|
|
|
|
|
|
|
|
self.log.info("Sending RBF transaction...")
|
|
|
|
utxo = self.wallet.get_utxo(mark_as_spent=True)
|
|
|
|
original_fee = Decimal(40000)
|
|
|
|
original_tx = self.wallet.send_self_transfer(
|
|
|
|
from_node=node, utxo_to_spend=utxo, fee=original_fee / COIN
|
|
|
|
)
|
|
|
|
entry_time = node.getmempoolentry(original_tx["txid"])["time"]
|
|
|
|
|
|
|
|
self.log.info("Sending replacement transaction...")
|
|
|
|
replacement_fee = Decimal(45000)
|
|
|
|
replacement_tx = self.wallet.send_self_transfer(
|
|
|
|
from_node=node, utxo_to_spend=utxo, fee=replacement_fee / COIN
|
|
|
|
)
|
|
|
|
|
|
|
|
self.log.info("Polling buffer...")
|
|
|
|
bpf.perf_buffer_poll(timeout=200)
|
|
|
|
|
|
|
|
self.log.info("Ensuring mempool:replaced event was handled successfully...")
|
2023-06-23 07:05:34 -04:00
|
|
|
assert_equal(1, len(events))
|
|
|
|
event = events[0]
|
Manage exceptions in bcc callback functions
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
2023-06-05 10:31:31 -04:00
|
|
|
assert_equal(bytes(event.replaced_hash)[::-1].hex(), original_tx["txid"])
|
|
|
|
assert_equal(event.replaced_vsize, original_tx["tx"].get_vsize())
|
|
|
|
assert_equal(event.replaced_fee, original_fee)
|
|
|
|
assert_equal(event.replaced_entry_time, entry_time)
|
|
|
|
assert_equal(bytes(event.replacement_hash)[::-1].hex(), replacement_tx["txid"])
|
|
|
|
assert_equal(event.replacement_vsize, replacement_tx["tx"].get_vsize())
|
|
|
|
assert_equal(event.replacement_fee, replacement_fee)
|
|
|
|
|
|
|
|
bpf.cleanup()
|
2023-03-29 07:17:29 -03:00
|
|
|
self.generate(self.wallet, 1)
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def rejected_test(self):
|
|
|
|
"""Create an invalid transaction and make sure the tracepoint returns
|
|
|
|
the expected txid, rejection reason, peer id, and peer address."""
|
|
|
|
|
2023-06-23 07:05:34 -04:00
|
|
|
events = []
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
self.log.info("Adding P2P connection...")
|
|
|
|
node = self.nodes[0]
|
|
|
|
node.add_p2p_connection(P2PDataStore())
|
|
|
|
|
|
|
|
self.log.info("Hooking into mempool:rejected tracepoint...")
|
|
|
|
ctx = USDT(pid=node.process.pid)
|
|
|
|
ctx.enable_probe(probe="mempool:rejected", fn_name="trace_rejected")
|
2023-10-09 10:38:37 -03:00
|
|
|
bpf = BPF(text=MEMPOOL_TRACEPOINTS_PROGRAM, usdt_contexts=[ctx], debug=0, cflags=["-Wno-error=implicit-function-declaration"])
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def handle_rejected_event(_, data, __):
|
2023-06-23 07:05:34 -04:00
|
|
|
events.append(bpf["rejected_events"].event(data))
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
bpf["rejected_events"].open_perf_buffer(handle_rejected_event)
|
|
|
|
|
|
|
|
self.log.info("Sending invalid transaction...")
|
|
|
|
tx = self.wallet.create_self_transfer(fee_rate=Decimal(0))
|
|
|
|
node.p2ps[0].send_txs_and_test([tx["tx"]], node, success=False)
|
|
|
|
|
|
|
|
self.log.info("Polling buffer...")
|
|
|
|
bpf.perf_buffer_poll(timeout=200)
|
|
|
|
|
|
|
|
self.log.info("Ensuring mempool:rejected event was handled successfully...")
|
2023-06-23 07:05:34 -04:00
|
|
|
assert_equal(1, len(events))
|
|
|
|
event = events[0]
|
Manage exceptions in bcc callback functions
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
2023-06-05 10:31:31 -04:00
|
|
|
assert_equal(bytes(event.hash)[::-1].hex(), tx["tx"].hash)
|
2023-07-17 07:54:02 -04:00
|
|
|
# The next test is already known to fail, so disable it to avoid
|
|
|
|
# wasting CPU time and developer time. See
|
|
|
|
# https://github.com/bitcoin/bitcoin/issues/27380
|
|
|
|
#assert_equal(event.reason.decode("UTF-8"), "min relay fee not met")
|
Manage exceptions in bcc callback functions
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
2023-06-05 10:31:31 -04:00
|
|
|
|
|
|
|
bpf.cleanup()
|
2023-03-29 07:17:29 -03:00
|
|
|
self.generate(self.wallet, 1)
|
Add mempool tracepoints
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
2022-11-18 09:56:46 -03:00
|
|
|
|
|
|
|
def run_test(self):
|
|
|
|
"""Tests the mempool:added, mempool:removed, mempool:replaced,
|
|
|
|
and mempool:rejected tracepoints."""
|
|
|
|
|
|
|
|
# Create some coinbase transactions and mature them so they can be spent
|
|
|
|
node = self.nodes[0]
|
|
|
|
self.wallet = MiniWallet(node)
|
|
|
|
self.generate(self.wallet, 4)
|
|
|
|
self.generate(node, COINBASE_MATURITY)
|
|
|
|
|
|
|
|
# Test individual tracepoints
|
|
|
|
self.added_test()
|
|
|
|
self.removed_test()
|
|
|
|
self.replaced_test()
|
|
|
|
self.rejected_test()
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
MempoolTracepointTest().main()
|