schedule snapshot snapshotInitial deliveries after each upgrade

[warner]

What is the Problem Being Solved?

Each time we start a new vat incarnation (either when the vat is first created, or just after each upgrade), we perform several expensive steps. The first is initialize-worker, which creates the worker, including SES lockdown, and the supervisor installation. The second is startVat, which creates and initializes liveslots, evaluates the entire userspace source bundle, and runs buildRootObject. Then, for contract vats, the very next delivery is startZcf, which retrieves and evaluates the entire contract bundle, and initializes (or re-initializes) the contract state.

From the start of the incarnation until the first heap-snapshot point, any worker restarts (due to host reboot, or vat eviction and subsequent reload) will have to repeat all these expensive steps. In practice, this can take a full second, mostly dominated by the startVat time. So we have the vat-warehouse to create a heap snapshot pretty early after startup, after a small number of deliveries, configured in a parameter named snapshotInitial. #7549 sets this value to 3, enough to get both startVat and startZcf included in the snapshot.

After the initial snapshot, snapshotInterval configures how frequently we make subsequent snapshots. Whenever the size of the current transcript span (everything since the last initialize-worker or load-snapshot) grows above snapshotInterval, we perform another snapshot.

However, the counters we use don't know about upgrades or incarnations. So we'll get an early snapshot on the first incarnation (because the absolute number of deliveries has exceeded snapshotInitial), but we don't get early snapshots on the second/etc incarnations.

Description of the Design

The fix will be to add a transcriptStore API that reveals information about the length of the current incarnation, maybe called transcriptStore.countIncarnationItems(vatID). I'm not sure about the best implementation, but it's nominally:

sqlite> SELECT COUNT(*) FROM transcriptItems WHERE vatID="v49" AND incarnation=(SELECT incarnation FROM transcriptSpans WHERE vatID="v49" AND isCurrent=1);

Then vat-warehouse should be configured to trigger a snapshot whenever countIncarnationItems > snapshotInitial, or when the length of the current span (i.e. everything that isn't in a snapshot) is greater than snapshotInterval.

Except of course the first condition should not keep making snapshots over and over again. So we should also check whether we have an active snapshot for our vatID at all: snapStore.getSnapshotInfo(vatID).

Security Considerations

none

Scaling Considerations

should improve worker reload time in the window after each upgrade, and before snapshotInterval commands a snapshot be taken

Test Plan

unit tests in test-transcript-entries.js

[warner]

This turned out to be easy to do in the process of fixing #7504, and didn't need an extra swingStore API. Instead it just asks "is there a current snapshot" and "how many entries are in the current transcript span". The first question tells us whether to apply snapshotInitial or snapshotInterval, against the count from the second.