mirror of
https://github.com/community-scripts/ProxmoxVED.git
synced 2026-02-25 05:57:26 +00:00
Extend telemetry reporting and ingestion
Add extended telemetry functions and server-side support: misc/api.func gains helpers (categorize_error, install timer, detect_gpu) and new reporters for tools, addons, and an extended post_update_to_api with duration, GPU and error_category. misc/data/service.go updated to accept and validate new fields (type: tool/addon, tool_name, parent_ct, gpu_vendor, gpu_passthrough, install_duration, error_category), expand allowed enums, include new fields in UpsertTelemetry and mapping, and add input sanitization. Also add telemetry-ingest.exe binary. These changes enable richer telemetry (tool/addon events, GPU info, durations and categorized errors) and server ingestion/validation for them.
This commit is contained in:
CanbiZ (MickLesk)
316
misc/api.func
316
misc/api.func
@@ -407,3 +407,319 @@ EOF
|
||||
|
||||
POST_UPDATE_DONE=true
|
||||
}
|
||||
|
||||
# ==============================================================================
|
||||
# SECTION 3: EXTENDED TELEMETRY FUNCTIONS
|
||||
# ==============================================================================
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# categorize_error()
|
||||
#
|
||||
# - Maps exit codes to error categories for better analytics
|
||||
# - Categories: network, storage, dependency, permission, timeout, config, resource, unknown
|
||||
# - Used to group errors in dashboard
|
||||
# ------------------------------------------------------------------------------
|
||||
categorize_error() {
|
||||
local code="$1"
|
||||
case "$code" in
|
||||
# Network errors
|
||||
6|7|22|28|35) echo "network" ;;
|
||||
|
||||
# Storage errors
|
||||
214|217|219) echo "storage" ;;
|
||||
|
||||
# Dependency/Package errors
|
||||
100|101|102|127|160|161|162) echo "dependency" ;;
|
||||
|
||||
# Permission errors
|
||||
126|152) echo "permission" ;;
|
||||
|
||||
# Timeout errors
|
||||
124|28|211) echo "timeout" ;;
|
||||
|
||||
# Configuration errors
|
||||
203|204|205|206|207|208) echo "config" ;;
|
||||
|
||||
# Resource errors (OOM, etc)
|
||||
137|134) echo "resource" ;;
|
||||
|
||||
# Default
|
||||
*) echo "unknown" ;;
|
||||
esac
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# start_install_timer()
|
||||
#
|
||||
# - Captures start time for installation duration tracking
|
||||
# - Call at the beginning of installation
|
||||
# - Sets INSTALL_START_TIME global variable
|
||||
# ------------------------------------------------------------------------------
|
||||
start_install_timer() {
|
||||
INSTALL_START_TIME=$(date +%s)
|
||||
export INSTALL_START_TIME
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# get_install_duration()
|
||||
#
|
||||
# - Returns elapsed seconds since start_install_timer() was called
|
||||
# - Returns 0 if timer was not started
|
||||
# ------------------------------------------------------------------------------
|
||||
get_install_duration() {
|
||||
if [[ -z "${INSTALL_START_TIME:-}" ]]; then
|
||||
echo "0"
|
||||
return
|
||||
fi
|
||||
local now=$(date +%s)
|
||||
echo $((now - INSTALL_START_TIME))
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# detect_gpu()
|
||||
#
|
||||
# - Detects GPU vendor and passthrough type
|
||||
# - Sets GPU_VENDOR and GPU_PASSTHROUGH globals
|
||||
# - Used for GPU analytics
|
||||
# ------------------------------------------------------------------------------
|
||||
detect_gpu() {
|
||||
GPU_VENDOR=""
|
||||
GPU_PASSTHROUGH="none"
|
||||
|
||||
# Detect Intel GPU
|
||||
if lspci 2>/dev/null | grep -qi "VGA.*Intel"; then
|
||||
GPU_VENDOR="intel"
|
||||
GPU_PASSTHROUGH="igpu"
|
||||
fi
|
||||
|
||||
# Detect AMD GPU
|
||||
if lspci 2>/dev/null | grep -qi "VGA.*AMD\|VGA.*ATI"; then
|
||||
GPU_VENDOR="amd"
|
||||
# Check if discrete
|
||||
if lspci 2>/dev/null | grep -qi "AMD.*Radeon"; then
|
||||
GPU_PASSTHROUGH="dgpu"
|
||||
else
|
||||
GPU_PASSTHROUGH="igpu"
|
||||
fi
|
||||
fi
|
||||
|
||||
# Detect NVIDIA GPU
|
||||
if lspci 2>/dev/null | grep -qi "VGA.*NVIDIA\|3D.*NVIDIA"; then
|
||||
GPU_VENDOR="nvidia"
|
||||
GPU_PASSTHROUGH="dgpu"
|
||||
fi
|
||||
|
||||
export GPU_VENDOR GPU_PASSTHROUGH
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# post_tool_to_api()
|
||||
#
|
||||
# - Reports tool usage to telemetry
|
||||
# - Arguments:
|
||||
# * $1: tool_name (e.g., "microcode", "lxc-update", "post-pve-install")
|
||||
# * $2: status ("success" or "failed")
|
||||
# * $3: exit_code (optional, default: 0 for success, 1 for failed)
|
||||
# - For PVE host tools, not container installations
|
||||
# ------------------------------------------------------------------------------
|
||||
post_tool_to_api() {
|
||||
command -v curl &>/dev/null || return 0
|
||||
[[ "${DIAGNOSTICS:-no}" == "no" ]] && return 0
|
||||
|
||||
local tool_name="${1:-unknown}"
|
||||
local status="${2:-success}"
|
||||
local exit_code="${3:-0}"
|
||||
local error="" error_category=""
|
||||
local uuid duration
|
||||
|
||||
# Generate UUID for this tool execution
|
||||
uuid=$(cat /proc/sys/kernel/random/uuid 2>/dev/null || uuidgen 2>/dev/null || echo "tool-$(date +%s)")
|
||||
duration=$(get_install_duration)
|
||||
|
||||
# Map status
|
||||
[[ "$status" == "done" ]] && status="success"
|
||||
|
||||
if [[ "$status" == "failed" ]]; then
|
||||
[[ ! "$exit_code" =~ ^[0-9]+$ ]] && exit_code=1
|
||||
error=$(explain_exit_code "$exit_code")
|
||||
error_category=$(categorize_error "$exit_code")
|
||||
fi
|
||||
|
||||
local pve_version=""
|
||||
if command -v pveversion &>/dev/null; then
|
||||
pve_version=$(pveversion 2>/dev/null | awk -F'[/ ]' '{print $2}') || true
|
||||
fi
|
||||
|
||||
local JSON_PAYLOAD
|
||||
JSON_PAYLOAD=$(cat <<EOF
|
||||
{
|
||||
"random_id": "${uuid}",
|
||||
"type": "tool",
|
||||
"nsapp": "${tool_name}",
|
||||
"tool_name": "${tool_name}",
|
||||
"status": "${status}",
|
||||
"exit_code": ${exit_code},
|
||||
"error": "${error}",
|
||||
"error_category": "${error_category}",
|
||||
"install_duration": ${duration:-0},
|
||||
"pve_version": "${pve_version}"
|
||||
}
|
||||
EOF
|
||||
)
|
||||
|
||||
curl -fsS -m "${TELEMETRY_TIMEOUT}" -X POST "${TELEMETRY_URL}" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d "$JSON_PAYLOAD" &>/dev/null || true
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# post_addon_to_api()
|
||||
#
|
||||
# - Reports addon installation to telemetry
|
||||
# - Arguments:
|
||||
# * $1: addon_name (e.g., "filebrowser", "netdata")
|
||||
# * $2: status ("success" or "failed")
|
||||
# * $3: parent_ct (optional, name of parent container)
|
||||
# * $4: exit_code (optional)
|
||||
# - For addons installed inside containers
|
||||
# ------------------------------------------------------------------------------
|
||||
post_addon_to_api() {
|
||||
command -v curl &>/dev/null || return 0
|
||||
[[ "${DIAGNOSTICS:-no}" == "no" ]] && return 0
|
||||
|
||||
local addon_name="${1:-unknown}"
|
||||
local status="${2:-success}"
|
||||
local parent_ct="${3:-}"
|
||||
local exit_code="${4:-0}"
|
||||
local error="" error_category=""
|
||||
local uuid duration
|
||||
|
||||
# Generate UUID for this addon installation
|
||||
uuid=$(cat /proc/sys/kernel/random/uuid 2>/dev/null || uuidgen 2>/dev/null || echo "addon-$(date +%s)")
|
||||
duration=$(get_install_duration)
|
||||
|
||||
# Map status
|
||||
[[ "$status" == "done" ]] && status="success"
|
||||
|
||||
if [[ "$status" == "failed" ]]; then
|
||||
[[ ! "$exit_code" =~ ^[0-9]+$ ]] && exit_code=1
|
||||
error=$(explain_exit_code "$exit_code")
|
||||
error_category=$(categorize_error "$exit_code")
|
||||
fi
|
||||
|
||||
# Detect OS info
|
||||
local os_type="" os_version=""
|
||||
if [[ -f /etc/os-release ]]; then
|
||||
os_type=$(grep "^ID=" /etc/os-release | cut -d= -f2 | tr -d '"')
|
||||
os_version=$(grep "^VERSION_ID=" /etc/os-release | cut -d= -f2 | tr -d '"')
|
||||
fi
|
||||
|
||||
local JSON_PAYLOAD
|
||||
JSON_PAYLOAD=$(cat <<EOF
|
||||
{
|
||||
"random_id": "${uuid}",
|
||||
"type": "addon",
|
||||
"nsapp": "${addon_name}",
|
||||
"status": "${status}",
|
||||
"parent_ct": "${parent_ct}",
|
||||
"exit_code": ${exit_code},
|
||||
"error": "${error}",
|
||||
"error_category": "${error_category}",
|
||||
"install_duration": ${duration:-0},
|
||||
"os_type": "${os_type}",
|
||||
"os_version": "${os_version}"
|
||||
}
|
||||
EOF
|
||||
)
|
||||
|
||||
curl -fsS -m "${TELEMETRY_TIMEOUT}" -X POST "${TELEMETRY_URL}" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d "$JSON_PAYLOAD" &>/dev/null || true
|
||||
}
|
||||
|
||||
# ------------------------------------------------------------------------------
|
||||
# post_update_to_api_extended()
|
||||
#
|
||||
# - Extended version of post_update_to_api with duration, GPU, and error category
|
||||
# - Same arguments as post_update_to_api:
|
||||
# * $1: status ("done" or "failed")
|
||||
# * $2: exit_code (numeric)
|
||||
# - Automatically includes:
|
||||
# * Install duration (if start_install_timer was called)
|
||||
# * Error category (for failed status)
|
||||
# * GPU info (if detect_gpu was called)
|
||||
# ------------------------------------------------------------------------------
|
||||
post_update_to_api_extended() {
|
||||
# Silent fail - telemetry should never break scripts
|
||||
command -v curl &>/dev/null || return 0
|
||||
|
||||
# Prevent duplicate submissions
|
||||
POST_UPDATE_DONE=${POST_UPDATE_DONE:-false}
|
||||
[[ "$POST_UPDATE_DONE" == "true" ]] && return 0
|
||||
|
||||
[[ "${DIAGNOSTICS:-no}" == "no" ]] && return 0
|
||||
[[ -z "${RANDOM_UUID:-}" ]] && return 0
|
||||
|
||||
local status="${1:-failed}"
|
||||
local raw_exit_code="${2:-1}"
|
||||
local exit_code=0 error="" pb_status error_category=""
|
||||
local duration gpu_vendor gpu_passthrough
|
||||
|
||||
# Get duration
|
||||
duration=$(get_install_duration)
|
||||
|
||||
# Get GPU info (if detected)
|
||||
gpu_vendor="${GPU_VENDOR:-}"
|
||||
gpu_passthrough="${GPU_PASSTHROUGH:-}"
|
||||
|
||||
# Map status to telemetry values
|
||||
case "$status" in
|
||||
done | success)
|
||||
pb_status="success"
|
||||
exit_code=0
|
||||
error=""
|
||||
error_category=""
|
||||
;;
|
||||
failed)
|
||||
pb_status="failed"
|
||||
;;
|
||||
*)
|
||||
pb_status="unknown"
|
||||
;;
|
||||
esac
|
||||
|
||||
# For failed/unknown status, resolve exit code and error description
|
||||
if [[ "$pb_status" == "failed" ]] || [[ "$pb_status" == "unknown" ]]; then
|
||||
if [[ "$raw_exit_code" =~ ^[0-9]+$ ]]; then
|
||||
exit_code="$raw_exit_code"
|
||||
else
|
||||
exit_code=1
|
||||
fi
|
||||
error=$(explain_exit_code "$exit_code")
|
||||
error_category=$(categorize_error "$exit_code")
|
||||
[[ -z "$error" ]] && error="Unknown error"
|
||||
fi
|
||||
|
||||
local JSON_PAYLOAD
|
||||
JSON_PAYLOAD=$(cat <<EOF
|
||||
{
|
||||
"random_id": "${RANDOM_UUID}",
|
||||
"type": "${TELEMETRY_TYPE:-lxc}",
|
||||
"nsapp": "${NSAPP:-unknown}",
|
||||
"status": "${pb_status}",
|
||||
"exit_code": ${exit_code},
|
||||
"error": "${error}",
|
||||
"error_category": "${error_category}",
|
||||
"install_duration": ${duration:-0},
|
||||
"gpu_vendor": "${gpu_vendor}",
|
||||
"gpu_passthrough": "${gpu_passthrough}"
|
||||
}
|
||||
EOF
|
||||
)
|
||||
|
||||
curl -fsS -m "${TELEMETRY_TIMEOUT}" -X POST "${TELEMETRY_URL}" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d "$JSON_PAYLOAD" &>/dev/null || true
|
||||
|
||||
POST_UPDATE_DONE=true
|
||||
}
|
||||
Reference in New Issue
Block a user