Question: Preloading. Cache is full.

[jdsika]

Hi team,

while developing the asam-osi-converter we encounter performance issues which are partly due to the implementation but I suspect that it might ALSO be related to Lichtblick itself. I would appreciate some hints to resolve this mystery for me :)

If I use Lichtblick (current master) and the asam-osi-converter v0.0.3. I am getting the following problem report:
Cache is full. Preloading of topics.

How and where is this preloading configured or decided over which time frame the preloading is done?
The mcap is to my knowledge allowing for the chunking of the data according to log_time in order to allow random access and decompression of the chunks. My expectation would be to adapt the preloading according to the replay speed?

Best regards
Carlo

[aneuwald-ctw]

Hi @jdsika,

To clarify, this is not about how data is accessed in MCAP. When referring to "Preload topics (messages)," it means that Lichtblick attempts to load all messages from a specified topic when required—such as when plotting a chart, where the complete dataset is necessary. This preloading is managed by the BlockLoader class which loads message blocks into memory and serves them to the relevant components.

Regarding capacity, the BlockLoader enforces a predefined limit. Each time it loads a block of messages, it verifies whether the accumulated size has reached the threshold, as seen in the following section of the code:

lichtblick/packages/suite-base/src/players/IterablePlayer/BlockLoader.ts

Lines 336 to 348 in 00ffb56

	if (totalBlockSizeBytes > this.#maxCacheSize) {
	this.#problemManager.addProblem("cache-full", {
	severity: "error",
	message: `Cache is full. Preloading for topics [${Array.from(
	topicsToFetch.keys(),
	).join(", ")}] has stopped on block ${currentBlockId + 1}/${this.#blocks.length}.`,
	tip: "Try reducing the number of topics that require preloading at a given time (e.g. in plots), or try to reduce the time range of the file.",
	});
	// We need to emit progress here so the player will emit a new state
	// containing the problem.
	progress(this.#calculateProgress(topics, totalBlockSizeBytes));
	return;
	}

The cache size limit is currently set to 1.0 GB, defined in:

lichtblick/packages/suite-base/src/players/IterablePlayer/IterablePlayer.ts

Line 56 in 00ffb56

const DEFAULT_CACHE_SIZE_BYTES = 1.0e9;

Please ensure that you are subscribing only to the necessary topics. If your MCAP file is large enough that preloading all messages from a topic exceeds this limit, consider optimizing your subscription strategy. If you're building from source, you can increase this limit, but do so at your own risk, as it may impact performance and memory usage.

Let me know if you need further details.