bitcoin/doc/tracing.md

User-space, Statically Defined Tracing (USDT) for Bitcoin Core

Bitcoin Core includes statically defined tracepoints to allow for more observability during development, debugging, code review, and production usage. These tracepoints make it possible to keep track of custom statistics and enable detailed monitoring of otherwise hidden internals. They have little to no performance impact when unused.

eBPF and USDT Overview
======================

                ┌──────────────────┐            ┌──────────────┐
                │ tracing script   │            │ bitcoind     │
                │==================│      2.    │==============│
                │  eBPF  │ tracing │      hooks │              │
                │  code  │ logic   │      into┌─┤►tracepoint 1─┼───┐ 3.
                └────┬───┴──▲──────┘          ├─┤►tracepoint 2 │   │ pass args
            1.       │      │ 4.              │ │ ...          │   │ to eBPF
    User    compiles │      │ pass data to    │ └──────────────┘   │ program
    Space    & loads │      │ tracing script  │                    │
    ─────────────────┼──────┼─────────────────┼────────────────────┼───
    Kernel           │      │                 │                    │
    Space       ┌──┬─▼──────┴─────────────────┴────────────┐       │
                │  │  eBPF program                         │◄──────┘
                │  └───────────────────────────────────────┤
                │ eBPF kernel Virtual Machine (sandboxed)  │
                └──────────────────────────────────────────┘

1. The tracing script compiles the eBPF code and loads the eBPF program into a kernel VM
2. The eBPF program hooks into one or more tracepoints
3. When the tracepoint is called, the arguments are passed to the eBPF program
4. The eBPF program processes the arguments and returns data to the tracing script

The Linux kernel can hook into the tracepoints during runtime and pass data to sandboxed eBPF programs running in the kernel. These eBPF programs can, for example, collect statistics or pass data back to user-space scripts for further processing.

The two main eBPF front-ends with support for USDT are bpftrace and BPF Compiler Collection (BCC). BCC is used for complex tools and daemons and bpftrace is preferred for one-liners and shorter scripts. Examples for both can be found in contrib/tracing.

Tracepoint documentation

The currently available tracepoints are listed here.

Context net

Tracepoint net:inbound_message

Is called when a message is received from a peer over the P2P network. Passes information about our peer, the connection and the message as arguments.

Arguments passed:

1. Peer ID as int64
2. Peer Address and Port (IPv4, IPv6, Tor v3, I2P, ...) as pointer to C-style String (max. length 68 characters)
3. Connection Type (inbound, feeler, outbound-full-relay, ...) as pointer to C-style String (max. length 20 characters)
4. Message Type (inv, ping, getdata, addrv2, ...) as pointer to C-style String (max. length 20 characters)
5. Message Size in bytes as uint64
6. Message Bytes as pointer to unsigned chars (i.e. bytes)

Note: The message is passed to the tracepoint in full, however, due to space limitations in the eBPF kernel VM it might not be possible to pass the message to user-space in full. Messages longer than a 32kb might be cut off. This can be detected in tracing scripts by comparing the message size to the length of the passed message.

Tracepoint net:outbound_message

Is called when a message is sent to a peer over the P2P network. Passes information about our peer, the connection and the message as arguments.

Arguments passed:

1. Peer ID as int64
2. Peer Address and Port (IPv4, IPv6, Tor v3, I2P, ...) as pointer to C-style String (max. length 68 characters)
3. Connection Type (inbound, feeler, outbound-full-relay, ...) as pointer to C-style String (max. length 20 characters)
4. Message Type (inv, ping, getdata, addrv2, ...) as pointer to C-style String (max. length 20 characters)
5. Message Size in bytes as uint64
6. Message Bytes as pointer to unsigned chars (i.e. bytes)

Note: The message is passed to the tracepoint in full, however, due to space limitations in the eBPF kernel VM it might not be possible to pass the message to user-space in full. Messages longer than a 32kb might be cut off. This can be detected in tracing scripts by comparing the message size to the length of the passed message.

Context validation

Tracepoint validation:block_connected

Is called after a block is connected to the chain. Can, for example, be used to benchmark block connections together with -reindex.

Arguments passed:

1. Block Header Hash as pointer to unsigned chars (i.e. 32 bytes in little-endian)
2. Block Height as int32
3. Transactions in the Block as uint64
4. Inputs spend in the Block as int32
5. SigOps in the Block (excluding coinbase SigOps) uint64
6. Time it took to connect the Block in microseconds (µs) as uint64

Context utxocache

The following tracepoints cover the in-memory UTXO cache. UTXOs are, for example, added to and removed (spent) from the cache when we connect a new block. Note: Bitcoin Core uses temporary clones of the main UTXO cache (chainstate.CoinsTip()). For example, the RPCs generateblock and getblocktemplate call TestBlockValidity(), which applies the UTXO set changes to a temporary cache. Similarly, mempool consistency checks, which are frequent on regtest, also apply the UTXO set changes to a temporary cache. Changes to the main UTXO cache and to temporary caches trigger the tracepoints. We can't tell if a temporary cache or the main cache was changed.

Tracepoint utxocache:flush

Is called after the in-memory UTXO cache is flushed.

Arguments passed:

1. Time it took to flush the cache microseconds as int64
2. Flush state mode as uint32. It's an enumerator class with values 0 (NONE), 1 (IF_NEEDED), 2 (PERIODIC), 3 (ALWAYS)
3. Cache size (number of coins) before the flush as uint64
4. Cache memory usage in bytes as uint64
5. If pruning caused the flush as bool

Tracepoint utxocache:add

Is called when a coin is added to a UTXO cache. This can be a temporary UTXO cache too.

Arguments passed:

1. Transaction ID (hash) as pointer to unsigned chars (i.e. 32 bytes in little-endian)
2. Output index as uint32
3. Block height the coin was added to the UTXO-set as uint32
4. Value of the coin as int64
5. If the coin is a coinbase as bool

Tracepoint utxocache:spent

Is called when a coin is spent from a UTXO cache. This can be a temporary UTXO cache too.

Arguments passed:

1. Transaction ID (hash) as pointer to unsigned chars (i.e. 32 bytes in little-endian)
2. Output index as uint32
3. Block height the coin was spent, as uint32
4. Value of the coin as int64
5. If the coin is a coinbase as bool

Tracepoint utxocache:uncache

Is called when a coin is purposefully unloaded from a UTXO cache. This happens, for example, when we load an UTXO into a cache when trying to accept a transaction that turns out to be invalid. The loaded UTXO is uncached to avoid filling our UTXO cache up with irrelevant UTXOs.

Arguments passed:

1. Transaction ID (hash) as pointer to unsigned chars (i.e. 32 bytes in little-endian)
2. Output index as uint32
3. Block height the coin was uncached, as uint32
4. Value of the coin as int64
5. If the coin is a coinbase as bool

Context coin_selection

Tracepoint coin_selection:selected_coins

Is called when SelectCoins completes.

Arguments passed:

1. Wallet name as pointer to C-style string
2. Coin selection algorithm name as pointer to C-style string
3. Selection target value as int64
4. Calculated waste metric of the solution as int64
5. Total value of the selected inputs as int64

Tracepoint coin_selection:normal_create_tx_internal

Is called when the first CreateTransactionInternal completes.

Arguments passed:

1. Wallet name as pointer to C-style string
2. Whether CreateTransactionInternal succeeded as bool
3. The expected transaction fee as an int64
4. The position of the change output as an int32

Tracepoint coin_selection:attempting_aps_create_tx

Is called when CreateTransactionInternal is called the second time for the optimistic Avoid Partial Spends selection attempt. This is used to determine whether the next tracepoints called are for the Avoid Partial Spends solution, or a different transaction.

Arguments passed:

1. Wallet name as pointer to C-style string

Tracepoint coin_selection:aps_create_tx_internal

Is called when the second CreateTransactionInternal with Avoid Partial Spends enabled completes.

Arguments passed:

1. Wallet name as pointer to C-style string
2. Whether the Avoid Partial Spends solution will be used as bool
3. Whether CreateTransactionInternal succeeded as bool
4. The expected transaction fee as an int64
5. The position of the change output as an int32

Adding tracepoints to Bitcoin Core

To add a new tracepoint, #include <util/trace.h> in the compilation unit where the tracepoint is inserted. Use one of the TRACEx macros listed below depending on the number of arguments passed to the tracepoint. Up to 12 arguments can be provided. The context and event specify the names by which the tracepoint is referred to. Please use snake_case and try to make sure that the tracepoint names make sense even without detailed knowledge of the implementation details. Do not forget to update the tracepoint list in this document.

#define TRACE(context, event)
#define TRACE1(context, event, a)
#define TRACE2(context, event, a, b)
#define TRACE3(context, event, a, b, c)
#define TRACE4(context, event, a, b, c, d)
#define TRACE5(context, event, a, b, c, d, e)
#define TRACE6(context, event, a, b, c, d, e, f)
#define TRACE7(context, event, a, b, c, d, e, f, g)
#define TRACE8(context, event, a, b, c, d, e, f, g, h)
#define TRACE9(context, event, a, b, c, d, e, f, g, h, i)
#define TRACE10(context, event, a, b, c, d, e, f, g, h, i, j)
#define TRACE11(context, event, a, b, c, d, e, f, g, h, i, j, k)
#define TRACE12(context, event, a, b, c, d, e, f, g, h, i, j, k, l)

For example:

TRACE6(net, inbound_message,
    pnode->GetId(),
    pnode->m_addr_name.c_str(),
    pnode->ConnectionTypeAsString().c_str(),
    sanitizedType.c_str(),
    msg.data.size(),
    msg.data.data()
);

Guidelines and best practices

Clear motivation and use case

Tracepoints need a clear motivation and use case. The motivation should outweigh the impact on, for example, code readability. There is no point in adding tracepoints that don't end up being used.

Provide an example

When adding a new tracepoint, provide an example. Examples can show the use case and help reviewers testing that the tracepoint works as intended. The examples can be kept simple but should give others a starting point when working with the tracepoint. See existing examples in contrib/tracing/.

No expensive computations for tracepoints

Data passed to the tracepoint should be inexpensive to compute. Although the tracepoint itself only has overhead when enabled, the code to compute arguments is always run - even if the tracepoint is not used. For example, avoid serialization and parsing.

Semi-stable API

Tracepoints should have a semi-stable API. Users should be able to rely on the tracepoints for scripting. This means tracepoints need to be documented, and the argument order ideally should not change. If there is an important reason to change argument order, make sure to document the change and update the examples using the tracepoint.

eBPF Virtual Machine limits

Keep the eBPF Virtual Machine limits in mind. eBPF programs receiving data from the tracepoints run in a sandboxed Linux kernel VM. This VM has a limited stack size of 512 bytes. Check if it makes sense to pass larger amounts of data, for example, with a tracing script that can handle the passed data.

bpftrace argument limit

While tracepoints can have up to 12 arguments, bpftrace scripts currently only support reading from the first six arguments (arg0 till arg5) on x86_64. bpftrace currently lacks real support for handling and printing binary data, like block header hashes and txids. When a tracepoint passes more than six arguments, then string and integer arguments should preferably be placed in the first six argument fields. Binary data can be placed in later arguments. The BCC supports reading from all 12 arguments.

Strings as C-style String

Generally, strings should be passed into the TRACEx macros as pointers to C-style strings (a null-terminated sequence of characters). For C++ std::strings, c_str() can be used. It's recommended to document the maximum expected string size if known.

Listing available tracepoints

Multiple tools can list the available tracepoints in a bitcoind binary with USDT support.

GDB - GNU Project Debugger

To list probes in Bitcoin Core, use info probes in gdb:

$ gdb ./src/bitcoind
…
(gdb) info probes
Type Provider   Name             Where              Semaphore Object
stap net        inbound_message  0x000000000014419e /src/bitcoind
stap net        outbound_message 0x0000000000107c05 /src/bitcoind
stap validation block_connected  0x00000000002fb10c /src/bitcoind
…

With readelf

The readelf tool can be used to display the USDT tracepoints in Bitcoin Core. Look for the notes with the description NT_STAPSDT.

$ readelf -n ./src/bitcoind | grep NT_STAPSDT -A 4 -B 2
Displaying notes found in: .note.stapsdt
  Owner                 Data size	Description
  stapsdt              0x0000005d	NT_STAPSDT (SystemTap probe descriptors)
    Provider: net
    Name: outbound_message
    Location: 0x0000000000107c05, Base: 0x0000000000579c90, Semaphore: 0x0000000000000000
    Arguments: -8@%r12 8@%rbx 8@%rdi 8@192(%rsp) 8@%rax 8@%rdx
…

With tplist

The tplist tool is provided by BCC (see Installing BCC). It displays kernel tracepoints or USDT probes and their formats (for more information, see the tplist usage demonstration). There are slight binary naming differences between distributions. For example, on Ubuntu the binary is called tplist-bpfcc.

$ tplist -l ./src/bitcoind -v
b'net':b'outbound_message' [sema 0x0]
  1 location(s)
  6 argument(s)
…