moonfire-nvr/server/db/schema.sql

427 lines
17 KiB
MySQL
Raw Normal View History

-- This file is part of Moonfire NVR, a security camera network video recorder.
-- Copyright (C) 2020 The Moonfire NVR Authors; see AUTHORS and LICENSE.txt.
-- SPDX-License-Identifier: GPL-v3.0-or-later WITH GPL-3.0-linking-exception.';
-- schema.sql: SQLite3 database schema for Moonfire NVR.
-- See also design/schema.md.
-- Database metadata. There should be exactly one row in this table.
create table meta (
2019-12-28 08:48:08 -05:00
uuid blob not null check (length(uuid) = 16),
-- Holds a json.GlobalConfig.
config text
);
-- This table tracks the schema version.
-- There is one row for the initial database creation (inserted below, after the
-- create statements) and one for each upgrade procedure (if any).
create table version (
id integer primary key,
-- The unix time as of the creation/upgrade, as determined by
-- cast(strftime('%s', 'now') as int).
unix_time integer not null,
-- Optional notes on the creation/upgrade; could include the binary version.
notes text
);
-- Tracks every time the database has been opened in read/write mode.
-- This is used to ensure directories are in sync with the database (see
2018-03-09 20:41:53 -05:00
-- schema.proto:DirMeta), to disambiguate uncommitted recordings, and
-- potentially to understand time problems.
create table open (
id integer primary key,
2018-03-09 20:41:53 -05:00
uuid blob unique not null check (length(uuid) = 16),
-- Information about when / how long the database was open. These may be all
-- null, for example in the open that represents all information written
-- prior to database version 3.
-- System time when the database was opened, in 90 kHz units since
-- 1970-01-01 00:00:00Z excluding leap seconds.
2018-03-09 20:41:53 -05:00
start_time_90k integer,
-- System time when the database was closed or (on crash) last flushed.
end_time_90k integer,
-- How long the database was open. This is end_time_90k - start_time_90k if
-- there were no time steps or leap seconds during this time.
duration_90k integer,
boot_uuid check (length(boot_uuid) = 16)
);
create table sample_file_dir (
id integer primary key,
uuid blob unique not null check (length(uuid) = 16),
2021-10-26 14:47:13 -04:00
-- See json.SampleFileDirConfig.
config text,
-- The last (read/write) open of this directory which fully completed.
-- See schema.proto:DirMeta for a more complete description.
last_complete_open_id integer references open (id)
);
create table camera (
id integer primary key,
uuid blob unique not null check (length(uuid) = 16),
-- A short name of the camera, used in log messages.
short_name text not null,
-- A serialized json.CameraConfig
config text not null
);
create table stream (
id integer primary key,
camera_id integer not null references camera (id),
sample_file_dir_id integer references sample_file_dir (id),
type text not null check (type in ('main', 'sub', 'ext')),
-- A serialized json.StreamConfig
config text not null,
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- The total number of recordings ever created on this stream, including
-- deleted ones. This is used for assigning the next recording id.
cum_recordings integer not null check (cum_recordings >= 0),
-- The total media duration of all recordings ever created on this stream.
cum_media_duration_90k integer not null check (cum_media_duration_90k >= 0),
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- The total number of runs (recordings with run_offset = 0) ever created
-- on this stream.
cum_runs integer not null check (cum_runs >= 0),
unique (camera_id, type)
);
-- Each row represents a single completed recorded segment of video.
-- Recordings are typically ~60 seconds; never more than 5 minutes.
create table recording (
-- The high 32 bits of composite_id are taken from the stream's id, which
-- improves locality. The low 32 bits are taken from the stream's
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- cum_recordings (which should be post-incremented in the same
-- transaction). It'd be simpler to use a "without rowid" table and separate
-- fields to make up the primary key, but
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- <https://www.sqlite.org/withoutrowid.html> points out that "without
-- rowid" is not appropriate when the average row size is in excess of 50
-- bytes. recording_cover rows (which match this id format) are typically
-- 1--5 KiB.
composite_id integer primary key,
-- The open in which this was committed to the database. For a given
-- composite_id, only one recording will ever be committed to the database,
-- but in-memory state may reflect a recording which never gets committed.
-- This field allows disambiguation in etags and such.
open_id integer not null references open (id),
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- This field is redundant with composite_id above, but used to enforce the
-- reference constraint and to structure the recording_start_time index.
stream_id integer not null references stream (id),
-- The offset of this recording within a run. 0 means this was the first
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- recording made from a RTSP session. The start of the run has composite_id
-- (composite_id-run_offset).
run_offset integer not null,
-- flags is a bitmask:
--
-- * 1, or "trailing zero", indicates that this recording is the last in a
-- stream. As the duration of a sample is not known until the next sample
-- is received, the final sample in this recording will have duration 0.
flags integer not null,
sample_file_bytes integer not null check (sample_file_bytes > 0),
-- The starting time of the recording, in 90 kHz units since
-- 1970-01-01 00:00:00 UTC excluding leap seconds. Currently on initial
-- connection, this is taken from the local system time; on subsequent
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- recordings in a run, it exactly matches the previous recording's end
-- time.
start_time_90k integer not null check (start_time_90k > 0),
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- The total duration of all previous recordings on this stream. This is
-- returned in API requests and may be helpful for timestamps in a HTML
-- MediaSourceExtensions SourceBuffer.
prev_media_duration_90k integer not null
check (prev_media_duration_90k >= 0),
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- The total number of previous runs (rows in which run_offset = 0).
prev_runs integer not null check (prev_runs >= 0),
-- The wall-time duration of the recording, in 90 kHz units. This is the
-- "corrected" duration.
wall_duration_90k integer not null
check (wall_duration_90k >= 0 and wall_duration_90k < 5*60*90000),
-- The media-time duration of the recording, relative to wall_duration_90k.
-- That is, media_duration_90k = wall_duration_90k + media_duration_delta_90k.
media_duration_delta_90k integer not null,
video_samples integer not null check (video_samples > 0),
2017-01-08 02:11:34 -05:00
video_sync_samples integer not null check (video_sync_samples > 0),
video_sample_entry_id integer references video_sample_entry (id),
-- The reason this run ended. Absent if there are more recordings in this
-- run or if this recording predates schema version 7.
end_reason text
check (composite_id >> 32 = stream_id)
);
create index recording_cover on recording (
-- Typical queries use "where stream_id = ? order by start_time_90k".
stream_id,
start_time_90k,
-- These fields are not used for ordering; they cover most queries so
-- that only database verification and actual viewing of recordings need
-- to consult the underlying row.
open_id,
wall_duration_90k,
media_duration_delta_90k,
video_samples,
video_sync_samples,
video_sample_entry_id,
sample_file_bytes,
run_offset,
flags
);
-- Fields which are only needed to check/correct database integrity problems
-- (such as incorrect timestamps).
create table recording_integrity (
2017-01-08 02:11:34 -05:00
-- See description on recording table.
composite_id integer primary key references recording (composite_id),
2017-01-08 02:11:34 -05:00
-- The number of 90 kHz units the local system's monotonic clock has
-- advanced more than the stated duration of recordings in a run since the
-- first recording ended. Negative numbers indicate the local system time is
-- behind the recording.
--
-- The first recording of a run (that is, one with run_offset=0) has null
-- local_time_delta_90k because errors are assumed to
-- be the result of initial buffering rather than frequency mismatch.
--
-- This value should be near 0 even on long runs in which the camera's clock
-- and local system's clock frequency differ because each recording's delta
-- is used to correct the durations of the next (up to 500 ppm error).
local_time_delta_90k integer,
-- The number of 90 kHz units the local system's monotonic clock had
-- advanced since the database was opened, as of the start of recording.
-- TODO: fill this in!
local_time_since_open_90k integer,
-- The difference between start_time_90k+duration_90k and a wall clock
-- timestamp captured at end of this recording. This is meaningful for all
-- recordings in a run, even the initial one (run_offset=0), because
-- start_time_90k is derived from the wall time as of when recording
-- starts, not when it ends.
-- TODO: fill this in!
wall_time_delta_90k integer,
-- The (possibly truncated) raw blake3 hash of the contents of the sample
-- file.
sample_file_blake3 blob check (length(sample_file_blake3) <= 32)
);
-- Large fields for a recording which are needed ony for playback.
-- In particular, when serving a byte range within a .mp4 file, the
-- recording_playback row is needed for the recording(s) corresponding to that
-- particular byte range, needed, but the recording rows suffice for all other
-- recordings in the .mp4.
create table recording_playback (
-- See description on recording table.
composite_id integer primary key references recording (composite_id),
2017-01-08 02:11:34 -05:00
-- See design/schema.md#video_index for a description of this field.
video_index blob not null check (length(video_index) > 0)
-- audio_index could be added here in the future.
);
-- Files which are to be deleted (may or may not still exist).
-- Note that besides these files, for each stream, any recordings >= its
track cumulative duration and runs This is useful for a combo scrub bar-based UI (#32) + live view UI (#59) in a non-obvious way. When constructing a HTML Media Source Extensions API SourceBuffer, the caller can specify a "mode" of either "segments" or "sequence": In "sequence" mode, playback assumes segments are added sequentially. This is good enough for a live view-only UI (#59) but not for a scrub bar UI in which you may want to seek backward to a segment you've never seen before. You will then need to insert a segment out-of-sequence. Imagine what happens when the user goes forward again until the end of the segment inserted immediately before it. The user should see the chronologically next segment or a pause for loading if it's unavailable. The best approximation of this is to track the mapping of timestamps to segments and insert a VTTCue with an enter/exit handler that seeks to the right position. But seeking isn't instantaneous; the user will likely briefly see first the segment they seeked to before. That's janky. Additionally, the "canplaythrough" event will behave strangely. In "segments" mode, playback respects the timestamps we set: * The obvious choice is to use wall clock timestamps. This is fine if they're known to be fixed and correct. They're not. The currently-recording segment may be "unanchored", meaning its start timestamp is not yet fixed. Older timestamps may overlap if the system clock was stepped between runs. The latter isn't /too/ bad from a user perspective, though it's confusing as a developer. We probably will only end up showing the more recent recording for a given timestamp anyway. But the former is quite annoying. It means we have to throw away part of the SourceBuffer that we may want to seek back (causing UI pauses when that happens) or keep our own spare copy of it (memory bloat). I'd like to avoid the whole mess. * Another approach is to use timestamps that are guaranteed to be in the correct order but that may have gaps. In particular, a timestamp of (recording_id * max_recording_duration) + time_within_recording. But again seeking isn't instantaneous. In my experiments, there's a visible pause between segments that drives me nuts. * Finally, the approach that led me to this schema change. Use timestamps that place each segment after the one before, possibly with an intentional gap between runs (to force a wait where we have an actual gap). This should make the browser's natural playback behavior work properly: it never goes to an incorrect place, and it only waits when/if we want it to. We have to maintain a mapping between its timestamps and segment ids but that's doable. This commit is only the schema change; the new data aren't exposed in the API yet, much less used by a UI. Note that stream.next_recording_id became stream.cum_recordings. I made a slight definition change in the process: recording ids for new streams start at 0 rather than 1. Various tests changed accordingly. The upgrade process makes a best effort to backfill these new fields, but of course it doesn't know the total duration or number of runs of previously deleted rows. That's good enough.
2020-06-09 19:17:32 -04:00
-- cum_recordings should be discarded on startup.
create table garbage (
-- This is _mostly_ redundant with composite_id, which contains the stream
-- id and thus a linkage to the sample file directory. Listing it here
-- explicitly means that streams can be deleted without losing the
-- association of garbage to directory.
sample_file_dir_id integer not null references sample_file_dir (id),
-- See description on recording table.
composite_id integer not null,
-- Organize the table first by directory, as that's how it will be queried.
primary key (sample_file_dir_id, composite_id)
) without rowid;
-- A concrete box derived from a ISO/IEC 14496-12 section 8.5.2
-- VisualSampleEntry box. Describes the codec, width, height, etc.
create table video_sample_entry (
id integer primary key,
-- The width and height in pixels; must match values within
-- `sample_entry_bytes`.
width integer not null check (width > 0),
height integer not null check (height > 0),
-- The codec in RFC-6381 format, such as "avc1.4d001f".
rfc6381_codec text not null,
-- The serialized box, including the leading length and box type (avcC in
-- the case of H.264).
data blob not null check (length(data) > 86),
-- Pixel aspect ratio, if known. As defined in ISO/IEC 14496-12 section
-- 12.1.4.
pasp_h_spacing integer not null default 1 check (pasp_h_spacing > 0),
pasp_v_spacing integer not null default 1 check (pasp_v_spacing > 0)
);
create table user (
id integer primary key,
username unique not null,
2021-10-26 16:08:45 -04:00
-- A json.UserConfig.
config text,
2021-10-26 16:08:45 -04:00
-- If set, a hash for password authentication, as generated by
-- `libpasta::hash_password`. This is separate from config for two reasons:
-- * It should never be sent over the wire, because password hashes are
-- almost as sensitive as passwords themselves. Keeping it separate avoids
-- complicating the protocol for retrieving the config and updating it
-- with optimistic concurrency control.
-- * It may be updated while authenticating to upgrade the password hash
-- format, and the conflicting writes again might complicate the update
-- protocol.
password_hash text,
-- A counter which increments with every password reset or clear.
password_id integer not null default 0,
-- Updated lazily on database flush; reset when password_id is incremented.
-- This could be used to automatically disable the password on hitting a threshold.
password_failure_count integer not null default 0,
-- Permissions available for newly created tokens or when authenticating via
-- unix_uid above. A serialized "Permissions" protobuf.
2021-10-26 16:08:45 -04:00
permissions blob not null default X''
);
-- A single session, whether for browser or robot use.
-- These map at the HTTP layer to an "s" cookie (exact format described
-- elsewhere), which holds the session id and an encrypted sequence number for
-- replay protection.
create table user_session (
-- The session id is a 48-byte blob. This is the unsalted Blake3 (32 bytes)
-- of the unencoded session id. Much like `password_hash`, a hash is used here
-- so that a leaked database backup can't be trivially used to steal
-- credentials.
session_id_hash blob primary key not null,
user_id integer references user (id) not null,
-- A 32-byte random number. Used to derive keys for the replay protection
-- and CSRF tokens.
seed blob not null,
-- A bitwise mask of flags, currently all properties of the HTTP cookie
-- used to hold the session:
-- 1: HttpOnly
-- 2: Secure
-- 4: SameSite=Lax
-- 8: SameSite=Strict - 4 must also be set.
flags integer not null,
-- The domain of the HTTP cookie used to store this session. The outbound
-- `Set-Cookie` header never specifies a scope, so this matches the `Host:` of
-- the inbound HTTP request (minus the :port, if any was specified).
domain text,
-- An editable description which might describe the device/program which uses
-- this session, such as "Chromebook", "iPhone", or "motion detection worker".
description text,
creation_password_id integer, -- the id it was created from, if created via password
creation_time_sec integer not null, -- sec since epoch
creation_user_agent text, -- User-Agent header from inbound HTTP request.
creation_peer_addr blob, -- IPv4 or IPv6 address, or null for Unix socket.
revocation_time_sec integer, -- sec since epoch
revocation_user_agent text, -- User-Agent header from inbound HTTP request.
revocation_peer_addr blob, -- IPv4 or IPv6 address, or null for Unix socket/no peer.
-- A value indicating the reason for revocation, with optional additional
-- text detail. Enumeration values:
-- 0: logout link clicked (i.e. from within the session itself)
-- 1: obsoleted by a change in hashing algorithm (eg schema 5->6 upgrade)
--
-- This might be extended for a variety of other reasons:
-- x: user revoked (while authenticated in another way)
-- x: password change invalidated all sessions created with that password
-- x: expired (due to fixed total time or time inactive)
-- x: evicted (due to too many sessions)
-- x: suspicious activity
revocation_reason integer,
revocation_reason_detail text,
-- Information about requests which used this session, updated lazily on database flush.
last_use_time_sec integer, -- sec since epoch
last_use_user_agent text, -- User-Agent header from inbound HTTP request.
last_use_peer_addr blob, -- IPv4 or IPv6 address, or null for Unix socket.
use_count not null default 0,
-- Permissions associated with this token; a serialized "Permissions" protobuf.
permissions blob not null default X''
) without rowid;
create index user_session_uid on user_session (user_id);
-- Timeseries with an enum value, eg:
-- * camera motion detection results (unknown, still, moving)
-- * security system arm status (unknown, disarmed, away, stay)
-- * security system zone status (unknown, normal, violated, trouble)
create table signal (
id integer primary key,
uuid blob unique not null check (length(uuid) = 16),
type_uuid blob not null references signal_type (uuid)
check (length(type_uuid) = 16),
-- Holds a json.SignalConfig
config text
);
create table signal_type (
uuid blob primary key check (length(uuid) = 16),
-- Holds a json.SignalTypeConfig
config text
) without rowid;
-- Changes to signals as of a given timestamp.
create table signal_change (
-- Event time, in 90 kHz units since 1970-01-01 00:00:00Z excluding leap seconds.
time_90k integer primary key,
-- Changes at this timestamp.
--
-- A blob of varints representing a list of
-- (signal number - next allowed, state) pairs, where signal number is
-- non-decreasing. For example,
-- input signals: 1 3 200 (must be sorted)
-- delta: 1 1 196 (must be non-negative)
-- states: 1 1 2
-- varint: \x01 \x01 \x01 \x01 \xc4 \x01 \x02
changes blob not null
);
insert into version (id, unix_time, notes)
values (7, cast(strftime('%s', 'now') as int), 'db creation');