[tarantool-patches] [PATCH v1 1/3] box: rfc for stacked diagnostic area in Tarantool
Kirill Shcherbatov
kshcherbatov at tarantool.org
Thu Aug 1 14:13:26 MSK 2019
Part of #1148
---
doc/rfc/1148-stacked-diagnostics.md | 136 ++++++++++++++++++++++++++++
1 file changed, 136 insertions(+)
create mode 100644 doc/rfc/1148-stacked-diagnostics.md
diff --git a/doc/rfc/1148-stacked-diagnostics.md b/doc/rfc/1148-stacked-diagnostics.md
new file mode 100644
index 000000000..a007491aa
--- /dev/null
+++ b/doc/rfc/1148-stacked-diagnostics.md
@@ -0,0 +1,136 @@
+# Stacked Diagnostics in Tarantool
+
+* **Status**: In progress
+* **Start date**: 30-08-2019
+* **Authors**: Kirill Shcherbatov @kshcherbatov kshcherbatov at tarantool.org, Konstantin Osipov @kostja kostja at tarantool.org, Georgy Kirichenko @GeorgyKirichenko georgy at tarantool.org, @Totktonada Alexander Turenko alexander.turenko at tarantool.org
+* **Issues**: [#1148](https://github.com/tarantool/<repository\>/issues/1148)
+
+## Summary
+Support stacked diagnostics for Tarantool allows to accumulate all occurred errors during processing a request. This allows to better understand what has happened and handle errors
+correspondingly.
+
+## Background and motivation
+
+Tarantool statements must produce diagnostic information that populates the diagnostics area. This is a Standard SQL requirement and other vendors and languages also have such feature.
+Diagnostics area stack must contain a diagnostics area for each nested execution context.
+
+### Current Tarantool's error diagnostics
+Currently Tarantool has `diag_set()` mechanism to set a diagnostic error.
+The last error is exported with `box.error.last()` endpoint.
+
+In total there are 5 error classes in Tarantool. Each error class is associated with set of error codes (@see ClientError::get_errcode(const struct error *e)).
+```
+ClientError -> ClientError::errcode() -> ...;
+OutOfMemory -> MEMORY_ISSUE
+SystemError -> SYSTEM
+CollationError -> CANT_CREATE_COLLATION
+Lua error -> PROC_LUA
+```
+All system-defined errors are exported with their unique error codes in box.error folder by NAME.
+
+You may use box.error.new endpoint to create a new error instance of predefined type:
+```
+tarantool> t = box.error.new(box.error.CREATE_SPACE, "myspace", "just I can't")
+tarantool> t:unpack()
+---
+- type: ClientError
+ code: 9
+ message: 'Failed to create space ''myspace'': just I can''t'
+ trace:
+ - file: '[string "t = box.error.new(box.error.CREATE_SPACE, "my..."]'
+ line: 1
+```
+
+User is allowed to define own errors with any code (including system errors) with
+```
+box.error.new{code = user_code, reason = user_error_msg}
+```
+where `user_code` must be greater that the biggest registered system error, say `10000`.
+
+Error cdata object has `:unpack()`, `:raise()`, `:match(...)`, `:__serialize()` methods and uniform `.type`, `.message` and `.trace` fields.
+
+
+## Proposal
+In some cases a diagnostic information must be more complicated.
+For example, when persistent Lua function referenced by functional index has a bug in it's definition, Lua handler sets an diag message. Then functional index extractor code sets an own, more specialized error.
+
+In this and similar situations, we need stack-based error diagnostics:
+
+- The serialize method should return the most recent error in stack:
+```
+tarantool> box.error.last()
+Failed to build a key for functional index ''idx'' of space ''withdata': function error
+```
+
+- The `:unpack()` method must return the complete diagnostics stack - a table of errors sorted in their historical order:
+```
+tarantool> box.error.last():unpack()
+---
+ - type: LuajitError
+ message: '[string "return function(tuple)..."]:2: attempt to call global
+ ''require'' (a nil value)'
+ trace:
+ - file: /home/kir/WORK/tarantool/src/lua/utils.c
+ line: 1000
+ - type: ClientError
+ code: 198
+ message: 'Failed to build a key for functional index ''idx'' of space ''withdata'':
+ function error
+ trace:
+ - file: /home/kir/WORK/tarantool/src/box/key_list.c
+ line: 68
+```
+
+- We need to introduce `next_error:wrap(error)` and `error:unwrap()` methods.
+The `wrap` function can take any previous and last error objects to append error to the end of next_error trace. The next_error backtrace is expected to be empty.
+The `unwrap` function removes the element's parent and returns the unwrapped content. This is effectively the inverse of the `wrap()` method.
+```
+Workaround:
+
+function create_new_user()
+ name, err = pcall(box.internal.func_call,
+ box.func.generate_user_name.name)
+ if not name then
+ local e = box.error.new({
+ code = APP_ERROR_CODE,
+ reason = "create_new_user"}):
+ e:wrap(err):raise()
+ end
+ box.schema.user.create(name)
+end
+```
+
+- We also need `error:has` method to test whether given error object contain given error code:
+```
+box.error.last():has(box.error.LuajitError)
+```
+
+## Detailed design
+
+Diagnostic area object has a good encapsulated API.
+
+Lets extend error object with a pointer to it's reason error and introduce a few new methods (we need to rename diag_add_error to diag_set_error because it replaces an existent diag error in fact):
+```
+struct error {
+...
+/* A pointer to the reason error. */
+ struct error *reason;
+};
+
+/**
+ * Add a new error to the diagnostics area: the previous error
+ * becomes a reason of a current.
+ * \param diag diagnostics area
+ * \param e error to add
+ */
+static inline void
+diag_add_error(struct diag *diag, struct error *e);
+
+/** Same as diag_set, but softly extend diagnostic information. */
+#define diag_add(class, ...)
+```
+
+### IPROTO Communication
+Currently errors are sent as (IPROTO_TYPE_ERROR | errcode) messages with error message string payload.
+Lets introduce a new IPROTO_TYPE_ERROR_TRACE error code with
+[[errcode1, "errstring1"], [errcode2, "errstring2"]..] payload, where errors are serialized in historical order.
--
2.22.0
More information about the Tarantool-patches
mailing list