A Babeltrace 2 sink.ctf.fs component writes the messages it
consumes to one or more CTF (see <https://diamon.org/ctf/>) 1.8
traces on the file system.
+-------------+
| sink.ctf.fs |
| +--> CTF trace(s) on
Messages -->@ in | the file system
+-------------+
See babeltrace2-intro(7) to learn more about the Babeltrace 2
project and its core concepts.
A sink.ctf.fs component does not merge traces: it writes the
messages of different input traces to different output traces.
Special trace IR to CTF translations
A sink.ctf.fs component makes a best effort to write CTF traces
that are semantically equivalent to the input traces. As of this
version, the component writes CTF 1.8 traces, so the following
field class translations can occur:
• The component translates a boolean field class to a CTF
unsigned 8-bit integer field class.
The unsigned integer field's value is 0 when the boolean
field's value is false and 1 when the boolean field's value
is true.
• The component translates a bit array field to a CTF unsigned
integer field class having the same length.
• The component translates an option field class to a CTF
variant field class where the options are an empty structure
field class and the optional field class itself.
The empty structure field is selected when the option field
has no field.
In all the cases above, the component adds a comment in the
metadata stream, above the field class, to indicate that a
special translation occurred.
Input message constraints
Because of limitations in CTF 1.8 regarding how discarded events
and packets are encoded:
• If a stream class supports discarded events and the ignore-
discarded-events parameter is NOT true:
• The stream class must support packets.
• Discarded events messages must have times.
• Any discarded events message must occur between a packet
end and a packet beginning message.
• The beginning time of a discarded events message must be
the same as the time of the last packet end message.
• The end time of a discarded events message must be the
same as the time of the next packet end message.
• Time ranges of discarded events messages must not
overlap.
• If a stream class supports discarded packets and the ignore-
discarded-packets parameter is NOT true:
• The stream class must support packets.
• Discarded packets messages must have times.
• The beginning time of a discarded events message must be
the same as the time of the last packet end message.
• The end time of a discarded events message must be the
same as the time of the next packet beginning message.
• Time ranges of discarded packets messages must not
overlap.
The messages which a source.ctf.fs component creates satisfy all
the requirements above.
If a discarded events or packets message has no events/packets
count, the sink.ctf.fs component adds 1 to the corresponding CTF
stream's counter.
Alignment and byte order
A sink.ctf.fs component always aligns data fields as such:
Integer fields with a size which is not a multiple of 8
1-bit.
All other scalar fields (integer, enumeration, real, string)
8-bit.
The component writes fields using the machine's native byte
order. As of this version, there's no way to force a custom byte
order.
Output path
The path of a CTF trace is the directory which directly contains
the metadata and data stream files.
The current strategy to build a path in which to write the
streams of a given input trace is, in this order:
1. If the assume-single-trace parameter is true, then the output
trace path to use for the single input trace is the directory
specified by the path parameter.
2. If the component recognizes the input trace as an LTTng (2.11
or greater) trace, then it checks specific trace environment
values to build a trace path relative to the directory
specified by the path parameter:
Linux kernel domain
HOST/SNAME-STIME/kernel
User space domain, per-UID buffering
HOST/SNAME-STIME/ust/uid/UID/ARCHW-bit
User space domain, per-PID buffering
HOST/SNAME-STIME/ust/pid/PNAME-PID-PTIME
With:
HOST
Target's hostname.
SNAME
Tracing session name.
STIME
Tracing session creation date/time.
UID
User ID.
ARCHW
Architecture's width (32 or 64).
PNAME
Process name.
PID
Process ID.
PTIME
Process's date/time.
3. If the input trace has a name, then the component sanitizes
this name and uses it as a relative path to the directory
specified by the path parameter.
The trace name sanitization operation:
• Replaces . subdirectories with _.
• Replaces .. subdirectories with __.
• Removes any trailing / character.
4. The component uses the subdirectory trace relative to the
directory specified by the path parameter.
In all the cases above, if the effective output trace path
already exists on the file system, the component appends a
numeric suffix to the name of the last subdirectory. The suffix
starts at 0 and increments until the path does not exist.
