logger: per-translation-unit channel filtering via CS_LOG_CHANNEL

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository

[akaitrade]

Per-channel logger filtering

Problem

Boost.Log's [Core] Filter is global. The node either logs everything at a chosen severity or it logs nothing. There is no way to raise debug verbosity
on one subsystem (e.g. consensus) without flooding the log file with every csdb / net / sync line.

Operators triaging a consensus issue had two choices: ship gigabytes of debug output or stay at info and lose the detail needed to diagnose.

Symptom

• One-subsystem investigations require a global severity bump
• Log files grow to tens of GB during longer debug sessions
• Channel-specific noise (e.g. csdb's per-call traces) masks the signal in unrelated investigations

Fix

Per-translation-unit channel tagging. Each .cpp file declares its channel at the top:

#defineCS_LOG_CHANNEL"consensus"
#include<lib/system/logger.hpp>

All bare csdebug() / cslog() / cserror() / etc. in that TU are then tagged with that channel name. Calls with an explicit logger type
(csdebug(logger::File)) keep their original behaviour.

Config gains a [log_channels] block:

[log_channels]default = info
consensus = debug
blockchain = warning
smartcontracts = info
sync = info
net = info

logger::initialize(settings, channelLevels) installs a lambda filter that consults the channel-level map per record, falling back to default then to
info when a channel is missing.

42 .cpp files tagged across csdb / csnode / net / solver:

• blockchain (19), consensus (14), net (6), smartcontracts (2), sync (1)

Verification

Compiles clean on Windows MSVC2022, Ubuntu 20.04 and Ubuntu 22.04. Files that pass an explicit logger type (logger::File, logger::None) are
unaffected; per-TU channel routing only applies to the bare convenience macros. Without a [log_channels] block the [Core] Filter path runs exactly
as before.

Risk

Low. Two compatibility surfaces:

• Existing configs without a [log_channels] block work unchanged (everything falls through to the global filter).
• Existing call sites with explicit logger types are routed via the original path, no behaviour change.

[cdmitry77]

To be discussed, logger is mainly used by developers, and including preliminary defined filters may cause often change of files group. Very good and useful cfunctionality byt it's better to be used by devs without commiting to main repository