I want to push the previous state (if the values changed) during an upsert
into an array history
(atomar) but no matter what I try, I can't find a clean solution.
Below you find simplified attempts:
Attempt 1: Use $set
and $concatArrays
final var timestamp = Document.parse(String.format("{$set: {timestamp: %d}}", time));
final var pointCount = Document.parse("{$set: {" +
"data_point_count: {$ifNull: [{$add: [\"$data_point_count\", 1]}, 1]} }}");
final var history = Document.parse(String.format("{$set: {" +
"history: {$cond: {" +
"if: {$eq: [\"$timestamp\", %d]}," +
"then: \"$history\"," + // no changes, keep untouched
"else: {$concatArrays: [" +
"{$ifNull: [\"$history\", []]}," + // without, history stays null
// Bug: this adds an empty object on 1st upsert (insert)
"[{timestamp: \"$timestamp\", data_point_count: \"$data_point_count\"}]" +
"]}" +
"}}" +
"}}", time));
final var update = Arrays.asList(setHistory, timestampUpdate, meanNinetyPercentileUpdateCommand, setDataPointCount);
// If I don't use Arrays.asList() commands like `$ifNull` are written into the database
//final var update = Updates.combine(setDataPointCount, setHistory); does not interpret the commands
final var options = new UpdateOptions().upsert(true);
targetCollection.updateOne(filter, update, options);
Problem:
upsert
is executed the first time (i.e. insert), an empty object is added to history: [{}]
[{}, {12345, 2}, {23456, 3}]
Attempt 2: Use $set
and $push
final var update = new Document();
final var set = new Document();
set.put("timestamp", time);
set.put("data_point_count", Document.parse("{$ifNull: [{$add: [\"$data_point_count\", 1]}, 1]}"));
update.put("$set", set);
update.put("$push", Document.parse("{history: {timestamp: \"$timestamp\", data_point_count: \"$data_point_count\"}}}"));
final var options = new UpdateOptions().upsert(true);
targetCollection.updateOne(filter, update, options);
Problem:
\"$timestamp\"
and $ifNull
and just writes e.g. $ifNull
into the database.Arrays.asList()
or else I get the error: Unrecognized pipeline stage name: '$push'
, but without the Arrays.asList
the commands are not executed but just written to the database, e.g. data_point_count: {$ifNull: ...}
Question:
How can I use one upsert
command which:
history: []
or history: null
or just no history
field during insertupdate
add the state before the update object to the history
array without any empty objects (see attempt 1).Update (suggestion from @Gibbs):
If I understood your suggestion correctly, it also adds an empty object when the upsert is executed the first time: [{}, {12345, 2}, {23456, 3}]
final var setHistory = Document.parse(String.format("{"+
"$set: {"+
"history: {"+
"$cond: ["+
// <Array condition goes here>
"{"+
"$eq: [\"$timestamp\", %d]"+ // with eq => insert also adds an empty object
//"$ne: [\"$timestamp\", %d]"+ // Error: "input to $map must be an array not string"
"},"+
//True part (timestamp not changed: keep history)
"{"+
"$map: {"+
"input: \"history\","+
"in: {"+
"$mergeObjects: ["+
"\"$$this\","+
"{"+
"timestamp: \"$timestamp\"," +
"mean_ninety_percentile: \"$mean_ninety_percentile\""+
"}"+
"]"+
"}"+
"}"+
"},"+
//False part (timestamp changed: add state to history)
"{"+
"$concatArrays: [" +
"{ $ifNull: [\"$history\", []] }," +
"[" +
"{"+
"timestamp: \"$timestamp\"," +
"mean_ninety_percentile: \"$mean_ninety_percentile\""+
"}"+
"]" +
"]"+
"}"+
"]"+
"}"+
"}"+
"}", time));
For clarification, the goal is the following state after the 2nd upsert
(i.e. 1 insert and 1 update):
collection = [
{
timestamp: 23456,
data_point_count: 2,
history: [
{timestamp: 12345, data_point_count: 1}
]
},...
]
You can achieve the same using aggregate update
. But it is supported from mongo 4.2+.
db.collection.update({},
[{
$set: {
timestamp: 12345,
mean_ninety_percentile: 0.111,
history: {
$cond: [
// <Array condition goes here>
{
$eq: [
"$timestamp",
12345
]
},
// True part (timestamp not changed: keep history)
"$history",
// False part (timestamp changed: add state to history)
{
$concatArrays: [
{
$ifNull: [
"$history",
[]
]
},
[
// add current state to history
{
timestamp: "$timestamp",
mean_ninety_percentile: "$mean_ninety_percentile"
}
]
}
]
}
}
}],
{
upsert: true
})