jiazhizhong/higress
1
0
Fork 0
mirror of https://github.com/alibaba/higress.git synced 2026-02-27 06:00:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: enhance model mapper and router with rebuild triggers and path extensions (#3218)

Browse Source
This commit is contained in:
澄潭
2025-12-12 18:10:57 +08:00
committed by GitHub
parent 5c17d3faa3
commit 0ada107ec5
6 changed files with 202 additions and 74 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
plugins/wasm-cpp/extensions/model_mapper/plugin.cc
View File

@@ -135,8 +135,40 @@ bool PluginRootContext::configure(size_t configuration_size) {
return true;
}
void PluginRootContext::incrementRequestCount() {
request_count_++;
if (request_count_ >= REBUILD_THRESHOLD) {
LOG_DEBUG("Request count reached threshold, triggering rebuild");
setFilterState("wasm_need_rebuild", "true");
request_count_ = 0; // Reset counter after setting rebuild flag
}
}
FilterHeadersStatus PluginRootContext::onHeader(
const ModelMapperConfigRule& rule) {
// Increment request count and check for rebuild
incrementRequestCount();
// Check memory threshold and trigger rebuild if needed
std::string value;
if (getValue({"plugin_vm_memory"}, &value)) {
// The value is stored as binary uint64_t, convert to string for logging
if (value.size() == sizeof(uint64_t)) {
uint64_t memory_size;
memcpy(&memory_size, value.data(), sizeof(uint64_t));
LOG_DEBUG(absl::StrCat("vm memory size is ", memory_size));
if (memory_size >= MEMORY_THRESHOLD_BYTES) {
LOG_INFO(absl::StrCat("Memory threshold reached (", memory_size, " >= ",
MEMORY_THRESHOLD_BYTES, "), triggering rebuild"));
setFilterState("wasm_need_rebuild", "true");
}
} else {
LOG_ERROR("invalid memory size format");
}
} else {
LOG_ERROR("get vm memory size failed");
}
if (!Wasm::Common::Http::hasRequestBody()) {
return FilterHeadersStatus::Continue;
}

11
plugins/wasm-cpp/extensions/model_mapper/plugin.h
View File

@@ -42,9 +42,10 @@ struct ModelMapperConfigRule {
std::vector<std::pair<std::string, std::string>> prefix_model_mapping_;
std::string default_model_mapping_;
std::vector<std::string> enable_on_path_suffix_ = {
"/completions", "/embeddings", "/images/generations",
"/audio/speech", "/fine_tuning/jobs", "/moderations",
"/image-synthesis", "/video-synthesis"};
"/completions", "/embeddings", "/images/generations",
"/audio/speech", "/fine_tuning/jobs", "/moderations",
"/image-synthesis", "/video-synthesis", "/rerank",
"/messages"};
};
// PluginRootContext is the root context for all streams processed by the
@@ -60,9 +61,13 @@ class PluginRootContext : public RootContext,
FilterHeadersStatus onHeader(const ModelMapperConfigRule&);
FilterDataStatus onBody(const ModelMapperConfigRule&, std::string_view);
bool configure(size_t);
void incrementRequestCount();
private:
bool parsePluginConfig(const json&, ModelMapperConfigRule&) override;
uint64_t request_count_ = 0;
static constexpr uint64_t REBUILD_THRESHOLD = 1000;
static constexpr size_t MEMORY_THRESHOLD_BYTES = 200 * 1024 * 1024;
};
// Per-stream context.