Elasticsearch node js point in time search_phase_execution_exception - node.js

Question

const body = {
query: {
geo_shape: {
geometry: {
relation: 'within',
shape: {
type: 'polygon',
coordinates: [$polygon],
},
},
},
},
pit: {
id: "t_yxAwEPZXNyaS1wYzYtMjAxN3IxFjZxU2RBTzNyUXhTUV9XbzhHSk9IZ3cAFjhlclRmRGFLUU5TVHZKNXZReUc3SWcAAAAAAAALmpMWQkNwYmVSeGVRaHU2aDFZZExFRjZXZwEWNnFTZEFPM3JReFNRX1dvOEdKT0hndwAA",
keep_alive: "1m",
},
};
Query fails with search_phase_execution_exception at onBody
Without pit query works fine but it's needed to retrieve more than 10000 hits

Answer 1

Well, using PIT in NodeJS ElasticSearch's client is not clear, or at least is not well documented. You can create a PIT using the client like:
const pitRes = await elastic.openPointInTime({
index: index,
keep_alive: "1m"
});
pit_id = pitRes.body.id;
But there is no way to use that pit_id in the search method, and it's not documented properly :S
BUT, you can use the scroll API as follows:
const scrollSearch = await elastic.helpers.scrollSearch({
index: index,
body: {
"size": 10000,
"query": {
"query_string": {
"fields": [ "vm_ref", "org", "vm" ],
"query": organization + moreQuery
},
"sort": [
{ "utc_date": "desc" }
]
}
}});
And then read the results as follows:
let res = [];
try {
for await (const result of scrollSearch) {
res.push(...result.body.hits.hits);
}
} catch (e) {
console.log(e);
}
I know that's not the exact answer to your question, but I hope it helps ;)

Answer 2

The usage of point-in-time for pagination of search results is now documented in ElasticSearch. You can find more or less detailed explanations here: Paginate search results
I prepared an example that may give an idea about how to implement the workflow, described in the documentation:
async function searchWithPointInTime(cluster, index, chunkSize, keepAlive) {
if (!chunkSize) {
chunkSize = 5000;
}
if (!keepAlive) {
keepAlive = "1m";
}
const client = new Client({ node: cluster });
let pointInTimeId = null;
let searchAfter = null;
try {
// Open point in time
pointInTimeId = (await client.openPointInTime({ index, keep_alive: keepAlive })).body.id;
// Query next chunk of data
while (true) {
const size = remained === null ? chunkSize : Math.min(remained, chunkSize);
const response = await client.search({
// Pay attention: no index here (because it will come from the point-in-time)
body: {
size: chunkSize,
track_total_hits: false, // This will make query faster
query: {
// (1) TODO: put any filter you need here (instead of match_all)
match_all: {},
},
pit: {
id: pointInTimeId,
keep_alive: keepAlive,
},
// Sorting should be by _shard_doc or at least include _shard_doc
sort: [{ _shard_doc: "desc" }],
// The next parameter is very important - it tells Elastic to bring us next portion
...(searchAfter !== null && { search_after: [searchAfter] }),
},
});
const { hits } = response.body.hits;
if (!hits || !hits.length) {
break; // No more data
}
for (hit of hits) {
// (2) TODO: Do whatever you need with results
}
// Check if we done reading the data
if (hits.length < size) {
break; // We finished reading all data
}
// Get next value for the 'search after' position
// by extracting the _shard_doc from the sort key of the last hit
searchAfter = hits[hits.length - 1].sort[0];
}
} catch (ex) {
console.error(ex);
} finally {
// Close point in time
if (pointInTime) {
await client.closePointInTime({ body: { id: pointInTime } });
}
}
}