What is best serialization format for boltdb for template data?

<p>If I wanted to store lists of data in boltdb that could be fed to html/template, what would be the best performing format? I wanted to avoid JSON reflect overhead if possible.</p> <hr/>**评论：**<br/><br/>THEHIPP0: <pre><p>I currently use the <a href="https://github.com/gogo/protobuf" rel="nofollow">faster implementation of Protocol buffers</a>, but you should check if <code>gob</code> or even <code>JSON</code> isn&#39;t fast enough for you.</p> <p>To clarify: I only use protobuf because instead of rendering templates I use <a href="http://grpc.io" rel="nofollow">gRPC</a> and had to write the protobuf defintions anyway.</p></pre>benbjohnson: <pre><p>I use gogoprotobuf for encoding when I use Bolt in applications. There are options to use the <code>unsafe</code> package which lets you avoid some allocations.</p> <p>I use gogoprotobuf for <a href="https://github.com/benbjohnson/scuttlebutt" rel="nofollow">this application</a> and my decoding time is about 12µs per object (and that converts from the internal protobuf type to an application type). Those benchmarks are coming from a small $5/mo DigitalOcean box too so it&#39;s not a beefy server.</p> <p>If you really need performance then you could also use a generated template language like <a href="https://github.com/benbjohnson/ego" rel="nofollow">ego</a>.</p></pre>inhies: <pre><p>JSON is great for debugging while building your application, gob will be faster though.</p></pre>elithrar_: <pre><p>encoding/json is nearly always faster than gob, if you are using once-off encoders—which is likely when encoding templates. gob is sometimes faster when you&#39;re using it as intended: for streaming encoding.</p></pre>inhies: <pre><p>I based my comment off of these benchmarks:</p> <p>BenchmarkGobEncode 1000000 2191 ns/op</p> <p>BenchmarkJsonEncode 500000 4738 ns/op </p> <p>BenchmarkGobDecode 1000000 2019 ns/op</p> <p>BenchmarkJsonDecode 200000 12993 ns/op</p> <p>From <a href="https://npf.io/2014/07/intro-to-boltdb-painless-performant-persistence/" rel="nofollow">https://npf.io/2014/07/intro-to-boltdb-painless-performant-persistence/</a></p></pre>elithrar_: <pre><p>encoding/json will be much faster than html/template, so I wouldn&#39;t discount it. template parsing is (relatively speaking) pretty slow due to escaping requirements. </p> <p>Alternatively, you could use something like ffjson (<a href="https://github.com/pquerna/ffjson" rel="nofollow">https://github.com/pquerna/ffjson</a>) to generate Marshal/Unmarshal methods on your structs and avoid reflection where possible, noting that reflection isn&#39;t always the performance killer it is perceived to be. As I touched on before, template rendering and DB access will be the bottlenecks. </p></pre>guesdo: <pre><p><a href="https://golang.org/pkg/encoding/gob/" rel="nofollow">https://golang.org/pkg/encoding/gob/</a></p> <p>That might help. Although I would just use JSON or BSON or MsgPack.</p></pre>seufert: <pre><p>I a Go only solution is sufficient then just use Go. It&#39;s the most convenient solution to use: No external description of your datastructure needed, best support for datatypes (afaik no time.Time on Protobuf for example) and the performance is much better than json. It&#39;s also quite conventient to add or remove fields.</p></pre>