Merge #1682

1682: Change the format of custom ranking rules when importing old dumps r=curquiza a=Kerollmops

This PR changes the format of the custom ranking rules from `asc(price)` to `title:asc` as the format changed between v0.21 and v0.22. The dumps are now correctly importing the custom ranking rules.

This PR also change the previous default ranking rules (without sort) to the new default ranking rules (with the new sort).

Co-authored-by: Kerollmops <clement@meilisearch.com>

meilisearch-http/src/index/dump.rs

@@ -8,7 +8,9 @@ use heed::RoTxn;
 use indexmap::IndexMap;
 use milli::update::{IndexDocumentsMethod, UpdateFormat::JsonStream};
 use serde::{Deserialize, Serialize};
+use serde_json::Value;
 
+use crate::index_controller::{asc_ranking_rule, desc_ranking_rule};
 use crate::option::IndexerOpts;
 
 use super::error::Result;
@@ -93,10 +95,22 @@ impl Index {
 
         let meta_path = src.as_ref().join(META_FILE_NAME);
         let mut meta_file = File::open(meta_path)?;
+
+        // We first deserialize the dump meta into a serde_json::Value and change
+        // the custom ranking rules settings from the old format to the new format.
+        let mut meta: Value = serde_json::from_reader(&mut meta_file)?;
+        if let Some(ranking_rules) = meta.pointer_mut("/settings/rankingRules") {
+            convert_custom_ranking_rules(ranking_rules);
+        }
+
+        // Then we serialize it back into a vec to deserialize it
+        // into a `DumpMeta` struct with the newly patched `rankingRules` format.
+        let patched_meta = serde_json::to_vec(&meta)?;
+
         let DumpMeta {
             settings,
             primary_key,
-        } = serde_json::from_reader(&mut meta_file)?;
+        } = serde_json::from_slice(&patched_meta)?;
         let settings = settings.check();
         let index = Self::open(&dst_dir_path, size)?;
         let mut txn = index.write_txn()?;
@@ -132,3 +146,25 @@ impl Index {
         Ok(())
     }
 }
+
+/// Converts the ranking rules from the format `asc(_)`, `desc(_)` to the format `_:asc`, `_:desc`.
+///
+/// This is done for compatibility reasons, and to avoid a new dump version,
+/// since the new syntax was introduced soon after the new dump version.
+fn convert_custom_ranking_rules(ranking_rules: &mut Value) {
+    *ranking_rules = match ranking_rules.take() {
+        Value::Array(values) => values
+            .into_iter()
+            .filter_map(|value| match value {
+                Value::String(s) if s.starts_with("asc") => asc_ranking_rule(&s)
+                    .map(|f| format!("{}:asc", f))
+                    .map(Value::String),
+                Value::String(s) if s.starts_with("desc") => desc_ranking_rule(&s)
+                    .map(|f| format!("{}:desc", f))
+                    .map(Value::String),
+                otherwise => Some(otherwise),
+            })
+            .collect(),
+        otherwise => otherwise,
+    }
+}

meilisearch-http/src/index_controller/dump_actor/loaders/v1.rs

@@ -12,6 +12,7 @@ use serde::{Deserialize, Deserializer, Serialize};
 use uuid::Uuid;
 
 use crate::index_controller::{self, uuid_resolver::HeedUuidStore, IndexMetadata};
+use crate::index_controller::{asc_ranking_rule, desc_ranking_rule};
 use crate::{
     index::{update_handler::UpdateHandler, Index, Unchecked},
     option::IndexerOpts,
@@ -164,19 +165,21 @@ impl From<Settings> for index_controller::Settings<Unchecked> {
                 None => Setting::NotSet
             },
             sortable_attributes: Setting::NotSet,
-            // we need to convert the old `Vec<String>` into a `BTreeSet<String>`
             ranking_rules: match settings.ranking_rules {
-                Some(Some(ranking_rules)) => Setting::Set(ranking_rules.into_iter().filter(|criterion| {
+                Some(Some(ranking_rules)) => Setting::Set(ranking_rules.into_iter().filter_map(|criterion| {
                     match criterion.as_str() {
-                        "words" | "typo" | "proximity" | "attribute" | "exactness" => true,
-                        s if s.starts_with("asc") || s.starts_with("desc") => true,
+                        "words" | "typo" | "proximity" | "attribute" | "exactness" => Some(criterion),
+                        s if s.starts_with("asc") => asc_ranking_rule(s).map(|f| format!("{}:asc", f)),
+                        s if s.starts_with("desc") => desc_ranking_rule(s).map(|f| format!("{}:desc", f)),
                         "wordsPosition" => {
-                            warn!("The criteria `attribute` and `wordsPosition` have been merged into a single criterion `attribute` so `wordsPositon` will be ignored");
-                            false
+                            warn!("The criteria `attribute` and `wordsPosition` have been merged \
+                                into a single criterion `attribute` so `wordsPositon` will be \
+                                ignored");
+                            None
                         }
                         s => {
                             error!("Unknown criterion found in the dump: `{}`, it will be ignored", s);
-                            false
+                            None
                         }
                     }
                 }).collect()),

meilisearch-http/src/index_controller/mod.rs

@@ -440,3 +440,17 @@ pub async fn get_arc_ownership_blocking<T>(mut item: Arc<T>) -> T {
         }
     }
 }
+
+/// Parses the v1 version of the Asc ranking rules `asc(price)`and returns the field name.
+pub fn asc_ranking_rule(text: &str) -> Option<&str> {
+    text.split_once("asc(")
+        .and_then(|(_, tail)| tail.rsplit_once(")"))
+        .map(|(field, _)| field)
+}
+
+/// Parses the v1 version of the Desc ranking rules `asc(price)`and returns the field name.
+pub fn desc_ranking_rule(text: &str) -> Option<&str> {
+    text.split_once("desc(")
+        .and_then(|(_, tail)| tail.rsplit_once(")"))
+        .map(|(field, _)| field)
+}