rarias/ovni
rarias/ovni
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Update trace specification to version 3

Browse Source
This commit is contained in:
Rodrigo Arias 2024-09-16 11:25:18 +02:00
parent bba46ac200
commit 73ce6ed035
1 changed files with 110 additions and 92 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

204
doc/user/runtime/trace_spec.md
View File

@ -1,127 +1,149 @@
# Trace specification
# Trace specification v3
!!! Important
This document refers to the trace specification for
the version 2
the version 3
The ovni instrumentation library stores the information collected in a
trace following the specification of this document.
The ovni instrumentation library libovni stores the information
collected in a runtime trace following the specification of this document.
## Structure
An ovni runtime trace (or simply, a trace) is composed of one or more
[streams](../concepts/trace-model.md#stream), which are directories containing
two mandatory files:
- `stream.json` the stream metadata in JSON format.
- `stream.obs` the binary stream with events.
Each stream is assigned to a single *part* in the [part
model](../concepts/part-model.md), usually assigned to a given thread.
There are no imposed rules on how to organize the several streams into
directories, but libovni uses the following approach for thread streams:
The complete trace is stored in a top-level directory named `ovni`.
Inside this directory you will find the loom directories with the prefix
`loom.`. The name of the loom is built from the `loom` parameter of
`ovni_proc_init()`, prefixing it with `loom.`.
Inside this directory you will find the loom directories. The name of
the loom directory is built from the `loom` parameter of `ovni_proc_init()`,
prefixing it with `loom.`.
Each loom directory contains one directory per process of that loom. The
name is composed of the `proc.` prefix and the PID of the process
specified in the `pid` argument to `ovni_proc_init()`.
Each process directory contains:
Inside each process there is one directory for each thread, composed by
the `thread.` prefix and the TID, which are the streams. The files
`stream.json` and `stream.obs` reside inside. Example:
- The process metadata file `metadata.json`.
- The thread streams, composed of:
- The binary stream like `thread.123.obs`
- The thread metadata like `thread.123.json`
```
ovni/loom.mio.nosv-u1000/proc.89719/thread.89719/stream.json
ovni/loom.mio.nosv-u1000/proc.89719/thread.89719/stream.obs
```
## Process metadata
This structure prevents collisions among threads with the same TID among nodes,
while allowing dumping events from a single thread, process or loom with
ovnidump.
!!! Important
## Stream metadata
Process metadata has version 2
The `stream.json` metadata file contains information about the part that
the stream is assigned to. This is generally used to determine the
hierarchy of the part model.
The process metadata file contains important information about the trace
that is invariant during the complete execution, and generally is
required to be available prior to processing the events in the trace.
The metadata is stored in the JSON file `metadata.json` inside each
process directory and contains the following keys:
The JSON must be an object (dictionary) with the following mandatory
keys:
- `version`: a number specifying the version of the metadata format.
Must have the value 2 for this version.
- `app_id`: the application ID, used to distinguish between applications
running on the same loom.
- `rank`: the rank of the MPI process (optional).
- `nranks`: number of total MPI processes (optional).
- `cpus`: the array of $`N_c`$ CPUs available in the loom. Only one
process in the loom must contain this mandatory key. Each element is a
dictionary with the keys:
- `index`: containing the logical CPU index from 0 to $`N_c - 1`$.
Must have the value 3 for this version.
The rest of information is stored for each model.
In particular, the `ovni` model enforces the use of:
- `ovni.part`: the type of part this stream is assigned to, usually
`thread`.
- `ovni.require`: a dictionary of model name and version which will
determine which models are enabled at emulation and the required
version.
- `ovni.finished`: must be 1 to ensure the stream is complete (mandatory
in all streams).
### Thread stream metadata
For `thread` streams, the following attributes are used.
- `ovni.tid`: the TID of the thread (mandatory, per-thread).
- `ovni.pid`: the PID of the process that the thread belongs to (mandatory, per-thread).
- `ovni.app_id`: the application ID of the process (optional, per-process).
- `ovni.rank`: the rank of the MPI process (optional, per-process).
- `ovni.nranks`: number of total MPI processes (optional, per-process).
- `ovni.loom`: the name of the loom that the process belongs to (mandatory, per-process).
- `ovni.loom_cpus`: the array of N CPUs available in the loom
(mandatory, per-loom). Each element is a dictionary with the keys:
- `index`: containing the logical CPU index from 0 to N - 1.
- `phyid`: the number of the CPU as given by the operating system
(which can exceed $`N_c`$).
(which can exceed N).
Here is an example of the `metadata.json` file:
Notice that some attributes don't need to be present in all thread
streams. For example, per-process requires that at least one thread
contains the attribute for each process. Similarly, per-loom requires
that at least one thread of the loom emits the attribute.
```
{
"version": 2,
"app_id": 1,
"rank": 0,
"nranks": 4,
"cpus": [
{
"index": 0,
"phyid": 0
},
{
"index": 1,
"phyid": 1
},
{
"index": 2,
"phyid": 2
},
{
"index": 3,
"phyid": 3
}
]
}
```
The final attribute value will be computed by merging all the values from the
children metadata. Simple values like numbers or strings must match exactly if
they appear duplicated, arrays are appended.
## Thread metadata
Other attributes can be used for other models.
!!! Important
Thread metadata has version 2
The thread metadata stores constant information per thread, like the
process metadata. The information is stored in a dictionary, where the
name of the emulation models are used as keys. In particular, the
libovni library writes information in the "ovni" key, such as the
model requirements, and other information like the version of libovni
used. Example:
Here is an example of the `stream.json` file for a thread of a nOS-V
program:
```json
{
"version": 2,
"version": 3,
"ovni": {
"lib": {
"version": "1.4.0",
"commit": "unknown"
"version": "1.10.0",
"commit": "dirty"
},
"part": "thread",
"tid": 89719,
"pid": 89719,
"loom": "mio.nosv-u1000",
"app_id": 1,
"require": {
"ovni": "1.0.0"
}
"ovni": "1.1.0",
"nosv": "2.3.0"
},
"loom_cpus": [
{ "index": 0, "phyid": 0 },
{ "index": 1, "phyid": 1 },
{ "index": 2, "phyid": 2 },
{ "index": 3, "phyid": 3 }
],
"finished": 1
},
"nosv": {
"can_breakdown": false,
"lib_version": "2.3.1"
}
}
```
The metadata is written to disk when the thread is first initialized
and when the thread finishes.
## Thread binary streams
## Binary stream
!!! Important
Thread binary stream has version 1
Binary streams have version 1
Streams are a binary files that contains a succession of events with
monotonically increasing clock values. Streams have a small header and
the variable size events just after the header.
A binary stream is a binary file named `stream.obs` that contains a
succession of events with monotonically increasing clock values. They
have a small header and the variable size events just after the header.
The header contains the magic 4 bytes of "ovni" and a version number of
4 bytes too. Here is a figure of the data stored in disk:
4 bytes too. Here is a figure of the data stored in disk on a little
endian machine:
![Stream](fig/stream.svg)
@ -145,7 +167,7 @@ payload:
- Normal events: with a payload up to 16 bytes
- Jumbo events: with a payload up to $`2^{32}`$ bytes
## Normal events
### Normal events
The normal events are composed of:
@ -178,7 +200,7 @@ In the following figure you can see each field annotated:
![Normal event with payload content](fig/event-normal-payload.svg)
## Jumbo events
### Jumbo events
The jumbo events are just like normal events but they can hold large
data. The size of the jumbo data is stored as a 32 bits integer as a
@ -203,10 +225,10 @@ In the following figure you can see each field annotated:
![Jumbo event](fig/event-jumbo.svg)
## Design considerations
### Design considerations
The stream format has been designed to be very simple, so writing a
parser library would take no more than 2 days for a single developer.
The binary stream format has been designed to be very simple, so writing
a parser library would take no more than 2 days for a single developer.
The size of the events has been designed to be small, with 12 bytes per
event when no payload is used.
@ -239,11 +261,7 @@ raw stream in binary, as the MCV codes can be read as ASCII characters:
This allows a human to detect signs of corruption by visually inspecting
the streams.
## Limitations
### Limitations
The streams are designed to be read only forward, as they only contain
the size of each event in the header.
Currently, we only support using the threads as sources of events, using
one stream per thread. However, adding support for more streams from
multiple sources is planned for the future.